VIETNAM-KOREA UNIVERSITY OF INFORMATION AND

COMMUNICATION TECHNOLOGY
FACULTY OF COMPUTER SCIENCE

GRADUATION THESIS
BUILDING A MANAGEMENT SYSTEM FOR A CHAIN OF
PRIVATE TRAINING AND STUDENT CARE CENTERS
USING AI TECHNOLOGY

Student: Hoang Ha
Class: 20SE6
Branch: Information Technology
Specialized: Software Engineer
Supervisor: Dr. Nguyen Van Binh

Da Nang, December - 2024

VIETNAM-KOREA UNIVERSITY OF INFORMATION AND
COMMUNICATION TECHNOLOGY
FACULTY OF COMPUTER SCIENCE

GRADUATION THESIS
BUILDING A MANAGEMENT SYSTEM FOR A CHAIN OF
PRIVATE TRAINING AND STUDENT CARE CENTERS
USING AI TECHNOLOGY

Student: Hoang Ha
Class: 20SE6
Branch: Information Technology
Specialized: Software Engineer
Supervisor: Dr. Nguyen Van Binh

Da Nang, December - 2024

 SUPERVISOR'S COMMENTS

……………………………………………………………………………………
……………………………………

……………………………………………………………………………………
……………………………………

……………………………………………………………………………………
……………………………………

……………………………………………………………………………………
……………………………………

……………………………………………………………………………………
……………………………………

……………………………………………………………………………………
……………………………………

……………………………………………………………………………………
……………………………………

……………………………………………………………………………………
……………………………………

……………………………………………………………………………………
……………………………………

……………………………………………………………………………………
……………………………………
Da Nang, December , 2024

Supervisor

Dr. Nguyen Van Binh

 ACKNOWLEDGEMENTS

First of all, with the deepest and most sincere feelings, please allow me to
express my gratitude to all individuals and organizations that have created favorable
conditions to support and help me throughout the process of studying and researching
this topic. During the time from the beginning of studying at the school until now, I
have received a lot of attention and help from teachers and friends.
With the deepest gratitude, I would like to send to the teachers at the Faculty of
Computer Science who have imparted valuable knowledge to us during my time
studying at the school. Thanks to the guidance and teachings of the teachers, my
research topic can be completed well.
Once again, I would like to sincerely thank Mr. Nguyen Van Binh - who has
directly helped, cared for, and guided me to successfully complete this report during
the past time.
The graduation thesis report was carried out in a period of nearly 3 months. My
initial steps into practice are still limited and confusing, so there are inevitably
shortcomings. I really hope to receive valuable comments from teachers so that my
knowledge in this field can be more complete and at the same time have the conditions
to supplement and raise my awareness.
Thank you very much!

Student,

Hoang Ha

i
 TABLE OF CONTENTS

ACKNOWLEDGEMENTS...........................................................................................i

TABLE OF CONTENTS..............................................................................................ii

LIST OF TABLES........................................................................................................iv

LIST OF FIGURES.......................................................................................................v

INTRODUCTION.........................................................................................................1

1. Problem statement...................................................................................................1

2. Aims and Objectives...............................................................................................1

3. Scope and objects of research.................................................................................2

4. Research methods...................................................................................................2

5. Practical significance of the topic...........................................................................3

6. Structure of the thesis..............................................................................................3

CHAPTER 1. TOPIC OVERVIEW............................................................................5

1.1 OVERVIEW.........................................................................................................5

1.1.1 Topic Overview.............................................................................................5

1.1.2 Trends in applying AI in education and current situation.............................5

1.2. METHODS AND EXPECTED RESULTS.........................................................6

1.2.1 Expected languages.......................................................................................6

1.2.2 Expected results.............................................................................................6

1.3. PLATFORM AND EXPECTED FUNCTIONALITY........................................7

1.3.1 Similar platform to central chain management system..................................7

1.3.2 Expected functional pages.............................................................................8

ii
 1.4. THEORETICAL FOUNDATIONS.....................................................................9

1.4.1 Laravel (PHP Framework).............................................................................9

1.4.2 MySQL Database Management System......................................................10

1.4.3 Python..........................................................................................................12

1.4.4 Docker..........................................................................................................13

1.4.5 Flask.............................................................................................................14

1.4.6 Chroma........................................................................................................16

1.4.7 Fast API.......................................................................................................17

1.4.8 Amazon Web Services.................................................................................18

1.5. CHAPTER 1 END.............................................................................................19

CHAPTER 2. SYSTEM ANALYSIS AND DESIGN...............................................20

2.1. OVERVIEW OF THE RESEARCH SYSTEM.................................................20

2.1.1 Functional requirements..............................................................................20

2.1.2 Non-functional requirements.......................................................................21

2.1.3 BFD functional hierarchy model.................................................................22

2.1.4 Use Case Diagram.......................................................................................23

2.1.5 Class diagram...............................................................................................30

2.1.6 Activity chart...............................................................................................32

2.1.7 Sequence diagram........................................................................................36

2.2. DATABASE......................................................................................................37

2.3. HOW RESTFUL APIS HANDLE DATA........................................................41

2.4. FACE DETECTION AND RECOGNITION....................................................42

2.4.1 Overview of face detection and recognition................................................42

iii
 2.4.2 Applying face detection and recognition to the system...............................43

2.5. APPLYING THE K-NEAREST NEIGHBORS MODEL TO ANALYZE AND

EVALUATE STUDENT BEHAVIOR....................................................................50

2.5.1 K-Nearest Neighbors Overview..................................................................50

2.5.2 Overview of the Student Stress Factors Dataset: A Comprehensive Analysis

..............................................................................................................................50

2.5.3 Make predictions on machine learning models...........................................51

2.6. APPLYING CONTENT-BASED RECOMMENDATION SYSTEMS TO THE

SYSTEM...................................................................................................................51

2.6.1 Overview of Content-based Recommendation Systems..............................51

2.6.2 Overview of the RAG (Retrieval-Augmented Generation) method............52

2.6.3 Working principle........................................................................................53

2.6.4 Overview of the keepitreal/vietnamese-sbert model...................................54

2.6.5 System architecture......................................................................................54

2.7 APPLY THE LLM MODEL TO BUILD CHATBOTS IN THE SYSTEM......55

2.7.1 Overview of chatbots...................................................................................55

2.7.2 Overview of the LLM model.......................................................................55

2.7.3 Operational process of chatbot in the system..............................................56

2.8 DEPLOY APIS TO AWS...................................................................................58

2.9 END OF CHAPTER 2........................................................................................58

CHAPTER 3. IMPLEMENTATION OF BUILDING SYSTEM...........................60

3.1 SYSTEM OVERVIEW......................................................................................60

3.2 RESULTS ACHIEVED......................................................................................60

3.2.1 About the management system for administrators and employees.............60

iv
 3.2.2 About the interface for parents and new users.............................................76

3.3 SUMMARY........................................................................................................88

CONCLUSIONS AND SUGGESTIONS..................................................................89

REFERENCES............................................................................................................vii

v
 LIST OF TABLES
Table 2.1 Student management function use case specification...................................27
Table 2.2 Student attendance use case specification.....................................................28
Table 2.3 Classroom management use case specification.............................................28
Table 2.4 Account table description..............................................................................36
Table 2.5 Center table description.................................................................................37
Table 2.6 Classroom table description..........................................................................37
Table 2.7 Compare the results of 2 facial recognition methods....................................47
Table 2.8 Comparing the prediction accuracy of 3 machine learning models..............49

vi
 LIST OF FIGURES
Figure 1.1 MISA EMIS system.......................................................................................7

Figure 1.2 EduForce System...........................................................................................7

Figure 1.3 Logo Laravel..................................................................................................9

Figure 1.4 Logo MySQL...............................................................................................10

Figure 1.5 How MySQL works.....................................................................................10

Figure 1.6 Logo Python.................................................................................................11

Figure 1.7 Logo Docker................................................................................................12

Figure 1.8 Logo Chroma...............................................................................................15

Figure 1.9 Logo FastAPI...............................................................................................16

Figure 1.10 Logo Amazon Web Services.....................................................................17

Figure 2.1 BFD functional hierarchy model.................................................................21

Figure 2.2 General level use case..................................................................................23

Figure 2.3 Teacher agent usecase..................................................................................25

Figure 2.4 Central management agent usecase.............................................................25

Figure 2.5 Parent agent usecase....................................................................................26

Figure 2.6 Student management function use case.......................................................26

Figure 2.7 Usecase of classroom management function...............................................27

Figure 2.8 Class diagram...............................................................................................29

Figure 2.9 Login Activity Diagram...............................................................................30

Figure 2.10 Activity diagram for adding new teachers.................................................31

Figure 2.11 Roll call activity diagram...........................................................................32

Figure 2.12 Chatbot Activity diagram...........................................................................33

vii
Figure 2.13 Login Sequence Diagram...........................................................................34

Figure 2.14 Sequence diagram adding new data...........................................................34

Figure 2.15 Database.....................................................................................................38

Figure 2.16 List API......................................................................................................39

Figure 2.17 API return result.........................................................................................39

Figure 2.18 Proposal Network......................................................................................41

Figure 2.19 P-Net CNN network...................................................................................41

Figure 2.20 R-Net..........................................................................................................42

Figure 2.21 O-Nets........................................................................................................42

Figure 2.22 Siamese Networks Architecture.................................................................43

Figure 2.23 Siamese Networks Architecture.................................................................44

Figure 2.24 FaceNet Algorithm....................................................................................44

Figure 2.25 Triplet loss function...................................................................................46

Figure 2.26 Loss function formula................................................................................46

Figure 2.27 Face recognition results.............................................................................47

Figure 2.28 Student Stress Factors Dataset...................................................................48

Figure 2.29 Prediction results of KNN model...............................................................49

Figure 2.30 Content-based Recommendation Systems.................................................50

Figure 2.31 RAG method..............................................................................................51

Figure 2.32 Recommendation Systems workflow........................................................52

Figure 2.33 Transformer Architecture...........................................................................54

Figure 2.34 Chatbot Activity Flow...............................................................................55

Figure 2.35 Deploy APIs to AWS.................................................................................56

viii
Figure 3.1 Role Management........................................................................................57

Figure 3.2 Account Management..................................................................................58

Figure 3.3 Class block management..............................................................................58

Figure 3.4 Semester Management.................................................................................59

Figure 3.5 Center Management.....................................................................................59

Figure 3.6 Human resource managementss...................................................................60

Figure 3.7 Admissions Management.............................................................................60

Figure 3.8 Contract Management..................................................................................61

Figure 3.9 Parental Management...................................................................................61

Figure 3.10 Student Management.................................................................................62

Figure 3.11 Contact parents..........................................................................................62

Figure 3.12 Email notification system..........................................................................63

Figure 3.13 Class Management.....................................................................................63

Figure 3.14 Subject Management..................................................................................64

Figure 3.15 Assignment Management..........................................................................64

Figure 3.16 Course Management..................................................................................65

Figure 3.17 Managing learning outcomes.....................................................................65

Figure 3.18 Student Health Management – 1................................................................66

Figure 3.19 Student Health Management – 2................................................................66

Figure 3.20 Student Health Management – 3................................................................67

Figure 3.21 Extracurricular Activities Management.....................................................67

Figure 3.22 Food Portion Management – 1...................................................................68

Figure 3.23 Food Portion Management – 2...................................................................68

ix
Figure 3.24 Food Management.....................................................................................69

Figure 3.25 Set up semester..........................................................................................69

Figure 3.26 Set up tuition..............................................................................................70

Figure 3.27 Login to the system....................................................................................70

Figure 3.28 Attendance screen......................................................................................71

Figure 3.29 Track attendance........................................................................................71

Figure 3.30 General statistics........................................................................................72

Figure 3.31 Revenue statistics.......................................................................................72

Figure 3.32 Stress level survey statistics.......................................................................73

Figure 3.33 Home..........................................................................................................73

Figure 3.34 Admissions Consulting..............................................................................74

Figure 3.35 Student Information – 1.............................................................................74

Figure 3.36 Student Information – 2.............................................................................75

Figure 3.37 Student Information – 3.............................................................................75

Figure 3.38 Student Information – 4.............................................................................75

Figure 3.39 Pay tuition fees via Momo portal...............................................................76

Figure 3.40 List of subjectss.........................................................................................76

Figure 3.41 Course details.............................................................................................77

Figure 3.42 Exercise details..........................................................................................77

Figure 3.43 Suggested exercises...................................................................................78

Figure 3.44 Instructions and comments on exercises....................................................78

Figure 3.45 Serving Size...............................................................................................79

Figure 3.46 Meal details................................................................................................79

x
Figure 3.47 Study Pressure Survey – 1.........................................................................80

Figure 3.48 Study Pressure Survey – 2.........................................................................80

Figure 3.49 Class schedule............................................................................................81

Figure 3.50 Extracurricular activities............................................................................81

Figure 3.51 Register for extracurricular activitiess.......................................................82

Figure 3.52 Chatbot interface........................................................................................82

Figure 3.53 Login page.................................................................................................83

Figure 3.54 About Us Page...........................................................................................83

Figure 3.55 Contact page..............................................................................................84

xi
Building a management system for a chain of private training and student care centers using AI technology

INTRODUCTION
1. Problem statement
In the current context, the demand for education and childcare at primary and
secondary levels is increasing, especially in large urban areas, where parents often do
not have much time to pick up and support their children in their studies as well as
extracurricular activities. Private training and childcare centers have become an
important part of the education system, providing not only academic knowledge but
also life skills, arts and sports. However, the effective management of this chain of
centers is facing many challenges, from monitoring students, managing teachers,
coordinating activities to ensuring service quality in a consistent and transparent
manner.
Traditional management systems, which are mainly based on paper or single
software, can no longer meet the practical requirements of current scale and
complexity. This leads to information loss, management errors, reduced performance
and affects the experience of parents and students. In this context, the application of
modern technology, especially artificial intelligence (AI), to build an integrated,
intelligent and efficient management system becomes an urgent solution. AI
technology not only helps automate management processes but also provides the
ability to analyze data, predict trends and make smart suggestions, contributing to
improving service quality and optimizing the operational efficiency of the chain of
centers.
In order to meet this need, I decided to choose the topic "Building a
management system for a chain of private training and student care centers using
AI technology" as the topic for my undergraduate thesis in Information Technology at
the Vietnam-Korea University of Information Technology and Communications. With
the desire to contribute to the application of modern technology in educational
management, this topic promises to bring practical and effective solutions, better
meeting the expectations of parents, students and society.
2. Aims and Objectives
This project aims to construct an efficient management system for private
student care and training centers using AI technology. The proposed system seeks to
address current challenges in managing operations across multiple centers, enhancing
efficiency and service quality while supporting comprehensive student care and
development. Our goal is to improve the user experience for administrators, teachers,

Hoang Ha – Class 20SE6 1

Building a management system for a chain of private training and student care centers using AI technology

parents, and students by integrating advanced technologies such as facial recognition,

behavioral monitoring, and learning analytics. The innovative system aims to provide
hassle-free procedures, including automated attendance tracking, scheduling, teaching
effectiveness analysis, and intelligent reporting. This breakthrough solution will help
private training centers optimize their operations, enhance the quality of student care
and education, and contribute to the sustainable development of the private education
sector.
3. Scope and objects of research
Geographical scope
- Private childcare and training centers in urban areas where there is a high
demand for educational and childcare services.
Object scope
- Primary and secondary school students studying at private training and childcare
centers
- Teachers and staff working at these centers
- Parents of students who need to monitor and manage their children's learning and
care process
Functional scope
The system will include functions for managing students, classes, teachers, staff,
invoices, services, subjects and learning outcomes.
Time range
Research and develop the system within a period of 03 months, including
requirements analysis, design, development, testing, implementation and evaluation stages
Technical scope
Ensure system security, performance and scalability to serve large numbers of
users and data.
4. Research methods
Document research methods
- Collect and analyze documents, scientific articles, and textbooks related to
educational management systems, AI technology in education, and existing student
management software solutions.

Hoang Ha – Class 20SE6 2

Building a management system for a chain of private training and student care centers using AI technology

- Learn about AI applications in the fields of facial recognition, automatic

attendance, and student assessment, supporting students and parents
System analysis method
- Analyze current educational management systems to assess the current
situation and determine actual needs at private training centers
- Provide reference models and propose solutions to integrate AI into the
management system
Comparison method
- Compare different AI technologies being applied in the field of education to
choose the most suitable technology for the system
- Evaluate the pros and cons of traditional management methods compared to
AI application solutions
5. Practical significance of the topic
The topic "Building a chain management system for private student
training and care centers applying AI technology" brings great practical
significance in the modern educational context. Applying AI helps automate
management processes, from student attendance, tuition management, to arranging
class schedules and reporting learning results, thereby minimizing errors and saving
time for students. centers. For parents, the system provides transparent and accurate
information about students' learning and living situation, helping to increase peace of
mind and trust in the center's services. On the student side, the application of modern
technology creates conditions for them to receive thoughtful care and study in a more
effective environment.
In addition, the project also contributes to improving management capacity for
private education centers, creating progress in digitizing and modernizing the
education sector, meeting development needs in the digital age. . With its expansion
potential, this system can become a standard management model widely applied, not
only at private training centers but also at other educational institutions in the future.
6. Structure of the thesis
After the Introduction, the thesis is structured in three chapters:
Chapter 1, Topic Overview This chapter marks a beginning of the thesis as it
provides an extensive background on the topic under consideration. This provides a
justification for the conceptualization, points out the limitations of the existing market

Hoang Ha – Class 20SE6 3

Building a management system for a chain of private training and student care centers using AI technology

solutions, and discusses pitfalls in the present-day implementations. Our system is

presented in this chapter as a proposed solution for solving identified complaints.
Furthermore, it also goes deep into the theoretical base, languages used and the means
of realization of the considered topic.
Chapter 2, System Analysis and Design The chapter outlines the findings from
the study that will provide the frameworks for action in order to create a useful
product. Overview of the employed tools and technologies, scrutiny of target
consumer group as well as requirements from user’s point of view. Moreover, the
chapter has elaborate explanations for use-case description, use-case diagram,
sequence diagram, activity diagram, among others, as well as database analysis.
Chapter 3, Implementation of Building System The next phase of this chapter
explains the complexities involved in stepwise analysis, looking at the selected
technology stacks, and providing an exemplary systematic model. It details the
architecture, user interface, sequence of operations, software construction, system
installation, etc.
Finally, there are Conclusions, Suggestions, References and Appendices related
to the topic.

Hoang Ha – Class 20SE6 4

Building a management system for a chain of private training and student care centers using AI technology

CHAPTER 1. TOPIC OVERVIEW

1.1 OVERVIEW
1.1.1 Topic Overview
With the continuous development of private education centers, effective
management of student, teacher, class, and financial information becomes increasingly
important to ensure service quality and response. increasing demands of parents. The
proposed system will take advantage of modern technologies, in which artificial
intelligence (AI) plays a supporting role, helping to automate management processes,
reduce manual work, and improve accuracy. precision in data processing. In addition,
the system also provides decision support tools, optimizes operational performance
and enhances user experience.
The topic not only brings practical benefits in improving management
efficiency but also contributes to promoting digital transformation in the private
education sector, towards building a smart, effective and sustainable management
system.
1.1.2 Trends in applying AI in education and current situation
In the context of the strong development of technology 4.0, the application of
technology in education has become an inevitable trend. Especially in private training
and student care centers, technology not only supports teaching but also plays an
important role in management, student care, and interaction with parents. Current
prominent trends include:
- Trends in applying technology in education:
+ Automate management processes: Many centers are moving from traditional
paper-based management methods and standalone software to integrated management
systems that automate tasks such as attendance, scheduling, and tuition management.
+ Artificial Intelligence (AI) Applications: AI is applied to support centers in many
areas such as monitoring student learning progress, predicting educational trends, and
personalizing learning programs.
+ Data digitization: Centers gradually convert all data from physical to digital
form, helping to minimize information loss and facilitate access and analysis.
- Current situation of private centers

Hoang Ha – Class 20SE6 5

Building a management system for a chain of private training and student care centers using AI technology

+ Growth in the number of centers: The need for education and care for students
outside of regular school hours is increasing, especially in large urban areas, causing
the number of private centers to continuously expand.
+ Difficulties in chain management: With centers operating in chains,
synchronizing information and monitoring service quality across the system is
difficult. Many centers still rely on traditional management methods, leading to data
inconsistencies and reduced operational efficiency.
+ Increasing demands from parents: Parents expect centers to not only provide
academic support but also provide comprehensive care, including health monitoring,
life skills development and support for extracurricular activities.
1.2. METHODS AND EXPECTED RESULTS
1.2.1 Expected languages
- About website:
+ Regarding the interface, we use the learned languages such as HTML, CSS,
JavaScript, and also use Bootstrap4 and jquery libraries to build a website interface.
+ Regarding system functions, the main language used is PHP and the Laravel
11.x Framework is also used.
- About AI:
+ Application of FaceNet model in face recognition
+ Applying KNN algorithm to predict students' stress level based on data from
StressLevelDataset.csv and survey information.
+ Using the RAG method and the keepitreal/vietnamese-sbert and Chroma
models to search and suggest similar exercises
+ Use the LLM (Large Language Model) model to implement the chatbot's
question answering function.
- MySQL database management system
- The system uses RESTful API to get data from the server.
1.2.2 Expected results
The result of the project is a complete private student care and training center
chain management system, integrating functions such as student management, class
schedules, automatic attendance using AI, and resource management. main and
statistical reports.

Hoang Ha – Class 20SE6 6

Building a management system for a chain of private training and student care centers using AI technology

1.3. PLATFORM AND EXPECTED FUNCTIONALITY

1.3.1 Similar platform to central chain management system
Currently, there are many education management systems and training centers
applying technology to optimize management activities, notably EduForce and MISA
EMIS.
MISA EMIS system: https://emis.misa.vn/
MISA EMIS is a comprehensive education management system designed to serve
schools and training centers. The system operates on a website platform with functions
such as student and teacher management, class schedules, attendance and financial
reports. MISA EMIS also supports the feature of connecting with parents through
notifications and reports on student learning situations. With an intuitive interface and
high customization capabilities, the system helps save management time and improve
the operational efficiency of educational centers.

Figure 1.1 MISA EMIS system

EduForce System: https://eduforce.omn1solution.com/

Hoang Ha – Class 20SE6 7

Building a management system for a chain of private training and student care centers using AI technology

Figure 1.2 EduForce System

EduForce is a professional education management platform that supports training
centers and schools in managing student information, class schedules, attendance, and
tuition fees. The system provides a user-friendly website interface, helping centers
easily organize and monitor teaching activities. EduForce also integrates features such
as reporting learning results statistics, sending notifications to parents, and financial
management, helping to simplify the operating process.
1.3.2 Expected functional pages
Drawing on experience from the above two systems, we plan to implement
functional pages.
- For administrators, center managers and center staff and teachers:
+ Login/ Forgot Password
+ Class management
+ Semester Management
+ Contract Management
+ Admissions Management
+ Central chain management
+ Account Management
+ Role Management
+ Human resource management

Hoang Ha – Class 20SE6 8

Building a management system for a chain of private training and student care centers using AI technology

+ Parental Management
+ Student Management
+ Classroom Management
+ Subject Management
+ Course Management
+ Managing learning outcomes
+ Student health management
+ Food Portion Management
+ Extracurricular Activities Management
- For parents and users:
+ Login/ Forgot Password
+ Home
+ Introduce
+ Register for admission
+ View student information
+ View course information
+ See suggested exercises
+ Register for extracurricular activities
+ View student's daily food intake
+ Monitor student stress levels through surveys
1.4. THEORETICAL FOUNDATIONS
1.4.1 Laravel (PHP Framework)
Concept:
- Laravel, as one of the most known PHP web frameworks in the MVC
approach, is free and open source. The development of Laravel was done by
Taylor Otwell24 and it aims in making the web development more simple
while still complying with the industry standard.

Hoang Ha – Class 20SE6 9

Building a management system for a chain of private training and student care centers using AI technology

Figure 1.3 Logo Laravel

Advantages:
Latest PHP features and a lot of documentation
- Compatibility with e-mail systems and fast turnaround time.
- Has a user friendly interface and is also safe with such things as PDO
preventing SQL injection.
- Designed according to the MVC based development.
Disadvantages:
- Updates of versions, destabilizing application consistency.
- This is very heavy on mobiles and may slow down loading web pages
Usage Model:
- It employs the MVC architecture separating Controller, Models, Views, thus
simplifying development and the design of a user interface. In addition, it helps
to maintain integrity in the programming principles.
1.4.2 MySQL Database Management System
Concept:
- Applications using open source database management system – MySQL,
working under the client-server model along with Apache and PHP. It is one of the
most popular RDBMS due to its features such as MySQL managing data through
databases and SQL. It was initially created by MySQL AB in 1994, with Oracle
Corporation acquiring it in 2008.

Hoang Ha – Class 20SE6 10

Building a management system for a chain of private training and student care centers using AI technology

Figure 1.4 Logo MySQL

- How MySQL Works:
The client-server model enables client computers to interact with the RDBMS
server by inputting SQL commands and requesting for the data. It can be used, for
instance, to edit rows in a table, add, remove or change the attributes of any object
inside a database. Big applications such as Facebook, Twitter, YouTube, Google and
Yahoo rely on MySQL for data storage

Figure 1.5 How MySQL works

Advantages:
- Flexibility and Ease of Use: This makes MySql’s install to be simple with no
hidden costs as source code modification is possible and free.
- High Performance: E-Commerce and technology business requires a robust
database management system that has lightning speed response irrespective of the
volume of data stored on it.
- Industry Standard: MySQL is a popular database that easily supports rapid
project rollout and hiring data scientists.
- Security: Safety of data is key in the industry and MySQL operates on high
security standards towards this direction.
Disadvantages:

Hoang Ha – Class 20SE6 11

Building a management system for a chain of private training and student care centers using AI technology

- Potential for Exploitation: Using MySQL for gaining control may be

exploitable.
- Limited in Massive Data Integration: Although it can manage a huge dataset;
MySQL is not very effective in handling massive and systematic data including the
massive supermarkets system and Population’s Information Management.
1.4.3 Python
Concept:
- A simple yet versatile high-level and object-oriented programming language –
Python was developed by Guido van Rossum.
- Using dynamic typing and automatic memory allocation, python is popular
due to its readability and wide applicability.

Figure 1.6 Logo Python

Advantages:
- It has simple grammar thus making it easy to study and comprehend.
- Supports itself by free and open source, with an active community.
- Portability across platforms.
- The extensibility that allows integrating code across other languages.
- Large standard library, high level interpreter language.
Disadvantages:
- Non-access modifiers such as protected, private, and public properties.
- Relatively slower processing speed than that of compiled languages.
- This results in higher memory consumption thus making it unsuitable
for constrained environments.

Hoang Ha – Class 20SE6 12

Building a management system for a chain of private training and student care centers using AI technology

- Unfit for applications like mobile and games development.

- Limited error detection during compilation.
1.4.4 Docker

Figure 1.7 Logo Docker

Concept:
- Docker is an open-source platform that offers containerization as a service for
deploying and scaling applications on flexible, portable containers. They enclose the
application together with its dependencies to ensure consistency across diverse
execution environments. Docker facilitates packaging of all the application related
requirements including libraries, runtime environment etc. in one container which runs
uniformly on any infrastructure, thereby making an identical computing environment
available wherever it gets deployed.
- How Docker Works: Application environments are created in a closed
environment by Docker using containerization technique. In essence, containers are an
application bundled with libraries and configurations into a standardized block.
Docker containers run under a shared OS kernel resulting in a lean infrastructure. With
Docker images as a blueprint, it becomes easy to share them anywhere with the ability
of developers to run applications consistently in a number of settings.
Advantages:
- Portability: Since docker containers are very portable, they work quite well in
diverse environments from development machines to production servers. It provides a
standard execution environment; thus, applications behave identically irrespective of
any environment; hence resolving the “it works on my machine” issue.

Hoang Ha – Class 20SE6 13

Building a management system for a chain of private training and student care centers using AI technology

- Isolation: Docker has isolated environments for applications, so dependencies will

not conflict with each other so applications will run normally without any interference
of processes or other systems on the hosting system.
- Rapid Deployment: Starting, stopping or scaling up their docker containers is
quite easy allowing for rapid application deployment. In dynamic and fast paced
development and deployment, this agility is quite useful.
- Resource Efficiency: The sharing of a single host’s kernel by docker
containers reduces the overall resource overhead in comparison to that of traditional
virtualization. As a result, it facilitates optimal usage of system resources which
increases the number of containers that can be ran in the same hardware.
- Microservices Architecture: Docker can be employed in a great way as it
offers developers the chance to design, deploy, and also scale different microservices.
The modular approach increases flexibility and scalability of larger and complicated
applications.
Disadvantages:
- Learning Curve: Developers who are not familiar with containerization
concepts and Docker specific commands would need to learn how to adopt docker.
Though the growing community and the comprehensive documentation counter this
problem.
- Security Concerns: The fact that Docker offers isolation of the environment
should not be a measure; some misconfiguration in some containers or security
vulnerabilities will always put them at risk. To reduce these risks, proper security
practices, regular updates and the application of best practices in containers are
paramount.
- Finally, Docker is a foundational part of current software engineering and
distribution methods. The portability, isolation, as well as rapid deployments which its
containerization technology provides make it an essential instrument for developing
scalable, consistent, and maintainable applications in variable computational settings.
1.4.5 Flask
Concept:
A simple, easy-to-use framework called Flask is used in developing a web
application on Python. Unlike some of the other monolithic frameworks though, Flask
adopts a microservices approach and only provides those components that are needed

Hoang Ha – Class 20SE6 14

Building a management system for a chain of private training and student care centers using AI technology

for each project or application it is being used for. It has attained fame because of its
simplicity, usage convenience, and scalability, meaning that many Python-based API
programs are created with it.
- How Flask Works: The approach used is straightforward and can be easily
modularized. It gives core needs of creating web applications but incurs no stringent
structures. This offers developers a chance to select more components as well as other
libraries depending on their requirements. It is possible to write different extensions
and applications using Flask since it well integrates with several of them.
Advantages:
- Simplicity and Flexibility: Flask has been built to be simple and lightweight
so that it can adapt to different situations in accordance with a developer’s own will. It
provides a framework approach to the creation of modules.
- Extensibility: The flexibility of a flask means that development of extensions
for functionality like authentication, integration with database, plus all other features is
easy. The flexibility expands the functional scope of the framework but without
introducing excessive complexity.
- Jinja Templating: Flask works with the Jinja template engine that assists in
easy production of the changing content that is used in web apps. This improves
development of UI, while providing easy integration with the end technologies.
- Werkzeug Integration: With regards to this point, flask utilizes Werkzeug
wsgi engine, which is quite robust in this regard. The integration into the framework
makes it strong as well as efficient.32
- Community and Documentation: There is a large support network in the
community of flask, as well as extensive literature about this platform that could be
helpful in the creation of applications by programmers.
Disadvantages:
- Opinionated Structure: The flexible design of Flask may appear advantageous;
however, it may present challenges in the development of huge projects. This is a
complex process that calls for great planning and organizing.
- Batteries Not Included: Flask is built on the motto “Batteries are not
included,” providing core features which developers should supplement with other
libraries for some features. However, some developers might opt for richer
frameworks.

Hoang Ha – Class 20SE6 15

Building a management system for a chain of private training and student care centers using AI technology

Finally, Flask is an all-purpose web framework for Python providing ease of

use as well as adaptability for programmers. It exhibits microservices architecture,
highly modular design, and significant community assistance. As such, this stack is
ideal for web apps and APIs that are highly customized and extensible in nature.
1.4.6 Chroma
Concept:
Chroma is a platform for storing and managing vector embeddings, which are
numerical representations of data that help machines understand and process text,
images, and other information. It enables fast and efficient retrieval of embeddings,
making it essential for applications like natural language processing (NLP), image
recognition, and recommendation systems.
- How Chroma Works: Chroma stores embeddings in a high-performance
database optimized for Approximate Nearest Neighbors (ANN) searches. This allows
rapid identification of similar embeddings, enabling real-time applications such as
semantic search and personalized recommendations.

Figure 1.8 Logo Chroma

Advantages:
- Fast Querying: Optimized for high-speed searches, supporting real-time
responses.
- Scalability: Capable of handling large datasets without significant
performance loss.
- Independent Management: Supports multiple embedding spaces for different
projects.

Hoang Ha – Class 20SE6 16

Building a management system for a chain of private training and student care centers using AI technology

- Wide Applications: Suitable for chatbots, recommendation systems, and

semantic search.
Disadvantages:
- Learning Curve: Requires understanding embeddings and vector databases to
use effectively.
- Resource Intensive: Demands substantial hardware resources for large-scale
deployments.
- Security Risks: Needs robust access control to prevent unauthorized access to
embeddings.
1.4.7 Fast API
Concept:
FastAPI is a modern Python web framework designed for creating APIs with
high performance and efficiency. It uses Python type hints to enable automatic data
validation, serialization, and interactive API documentation, making API development
faster and more reliable. FastAPI is built on Starlette and Pydantic, ensuring speed and
flexibility for both small-scale and enterprise-level applications.
- How FastAPI Works: FastAPI processes requests and responses using
Python’s type hints, which provide built-in data validation and type checking. It
supports both synchronous and asynchronous programming, enabling developers to
handle I/O-bound tasks efficiently. Interactive documentation is automatically
generated with tools like Swagger UI and ReDoc, streamlining testing and
communication with API consumers.

Figure 1.9 Logo FastAPI

Advantages:
- Offers high performance comparable to Node.js and Go, making it suitable for
demanding applications.
- Provides built-in data validation and type checking, reducing errors and
ensuring data consistency.

Hoang Ha – Class 20SE6 17

Building a management system for a chain of private training and student care centers using AI technology

- Automatically generates interactive API documentation, simplifying testing

and communication.
- Fully supports asynchronous programming, improving scalability for I/O-
bound tasks.
- Features an intuitive syntax and tools that speed up the development of
complex APIs.
1.4.8 Amazon Web Services

Figure 1.10 Logo Amazon Web Services

Concept:
Amazon Web Services (AWS) is a comprehensive cloud computing platform
that offers a wide range of services, including computing power, storage, and
databases, on a pay-as-you-go basis. AWS provides the infrastructure and tools needed
to build, deploy, and manage applications, making it essential for businesses of all
sizes to scale and innovate rapidly.
Advantages:
- Scalability: Easily scale resources up or down based on demand, ensuring cost
efficiency and performance
- Global Reach: AWS has data centers across the world, enabling low-latency
services and high availability
- Wide Range of Services: Offers over 200 fully-featured services, including
machine learning, IoT, analytics, and more
- Security: Provides robust security measures, including encryption, firewalls,
and compliance certifications
- Cost-Effective: Pay-as-you-go pricing model reduces upfront costs and allows
for flexible budgeting
Disadvantages:

Hoang Ha – Class 20SE6 18

Building a management system for a chain of private training and student care centers using AI technology

- Complexity: The vast number of services and configurations can be

overwhelming for beginners
- Cost Management: Without proper monitoring, costs can escalate quickly,
especially for large deployments
- Vendor Lock-In: Migrating away from AWS can be challenging and costly
due to dependencies on proprietary services
- Technical Knowledge: Requires expertise to fully leverage AWS's capabilities
and optimize performance
1.5. CHAPTER 1 END
This chapter provides us with a foundational knowledge to develop such
systems like programming tools, programming languages, and libraries and
frameworks. Know the ideas, terminologies, strengths and weaknesses of each one of
these tools, languages, methods and the procedure utilized by the algorithms.

Hoang Ha – Class 20SE6 19

Building a management system for a chain of private training and student care centers using AI technology

CHAPTER 2. SYSTEM ANALYSIS AND DESIGN

2.1. OVERVIEW OF THE RESEARCH SYSTEM
2.1.1 Functional requirements
The management system of a chain of private training and student care centers
includes the following functions
For parents and users:
+ Login/ Forgot Password
+ Home
+ Introduce
+ Register for admission
+ View student information
+ View course information
+ See suggested exercises
+ Register for extracurricular activities
+ View student's daily food intake
+ Monitor student stress levels through surveys
For administrators, center managers, teachers and center staff:
+ Login/ Forgot Password
+ Class management
+ Semester Management
+ Contract Management
+ Admissions Management
+ Central chain management
+ Account Management
+ Role Management
+ Human resource management
+ Parental Management
+ Student Management

Hoang Ha – Class 20SE6 20

Building a management system for a chain of private training and student care centers using AI technology

+ Classroom Management
+ Subject Management
+ Course Management
+ Managing learning outcomes
+ Student health management
+ Food Portion Management
+ Extracurricular Activities Management
2.1.2 Non-functional requirements
2.1.2.1 Performance
The system must process requests within 2 seconds, face recognition speed
under 1 second and handle up to 500 concurrent users without service interruption.
2.1.2.2 Scalability
The system needs to support data management for up to 1 million students and
easily expand when adding new centers without reducing performance.
2.1.2.3 Security
Data must be encrypted using SSL/TLS, GDPR compliant. The system needs to
clearly decentralize and only store vector embeddings instead of the original image.
2.1.2.4 Availability
The system ensures 99.9% availability, automatically backs up data daily and
restores within 30 minutes when a problem occurs.
2.1.2.5 Compatibility
The system must be compatible with devices such as computers and phones and
operate stably on popular browsers such as Chrome, Firefox and Safari.
2.1.2.6 Maintainability
The system needs to be easy to maintain and upgrade without service
interruption and have detailed documentation for users.
2.1.2.7 Reliability
Data must be 100% accurate, especially during face recognition and storage.
The system needs to automatically detect and handle minor errors.

Hoang Ha – Class 20SE6 21

Building a management system for a chain of private training and student care centers using AI technology

2.1.3 BFD functional hierarchy model

Figure 2.11 BFD functional hierarchy model

Hoang Ha – Class 20SE6 22

Building a management system for a chain of private training and student care centers using AI technology

2.1.4 Use Case Diagram

2.1.4.1 Identify the factors
- Use case diagram: there are objects that use this system
+ Administrator
+ Center Management
+ Staff
+ Teacher
+ Parents
2.1.4.2 The agents
- Administrator: Login, logout, center management, grade management,
semester management, personnel management, classroom management, subject
management, student health management, extracurricular activities management
- Center Management: personnel management, classroom management, subject
management, student health management, extracurricular activities management,
attendance management
- Staff: classroom management, subject management, student health
management, extracurricular activities management, attendance management.
- Teachers: classroom management, learning outcomes management, homework
management
- Parents: register for admissions consultation, view student information, view
student meals, view student academic results, view homework, monitor student stress
through surveys

Hoang Ha – Class 20SE6 23

Building a management system for a chain of private training and student care centers using AI technology

2.1.4.3 General use case diagram

Figure 2.12 General level use case

2.1.4.4 General Use Case Specification
Use Case Name: Manage Accounts
Primary Actor: Administrator
Preconditions: The Administrator is logged into the system with permission to
manage accounts
Postconditions: Account information is updated, and relevant changes (e.g.,
role updates) are reflected in the system
Main Flow:
The Administrator initiates the "Manage Accounts" use case
The system displays a list of existing user accounts
The Administrator selects one of the following actions:
Create New Account: Enter and save account details
Update Account Information: Modify and save existing account details

Hoang Ha – Class 20SE6 24

Building a management system for a chain of private training and student care centers using AI technology

Change Account Role: Select and assign a new role (e.g., Staff) → System
validates and updates permissions
Delete Account: Remove an existing account → System confirms the deletion
The system reflects the changes successfully in the account list
Extensions:
Create New Account:
The system automatically sends an email with account credentials or relevant
information to the new user after saving
Change Account Role:
The system ensures the selected role is valid and updates permissions
accordingly
Use Case Name: Manage Registration Contracts
Primary Actor: Administrator
Preconditions: The Administrator is logged into the system with permission to
manage contracts
Postconditions: Contract information is updated, contracts can be renewed, and
a printable version of the contract is available
Main Flow:
The Administrator initiates the "Manage Registration Contracts" use case
The system displays a list of existing contracts
Create New Contract: Enter and save the contract details (e.g., center name,
start date, duration, pricing)
Renew Contract: Select an existing contract to extend → Enter new duration
→ System updates and reflects the renewal
Update Contract Information: Modify contract details → Save changes
Delete Contract: Remove a contract → System confirms deletion
Print Contract: Generate a printable version of the selected contract
The system reflects all changes successfully and ensures the updated contract
data is available
Extensions:

Hoang Ha – Class 20SE6 25

Building a management system for a chain of private training and student care centers using AI technology

Renew Contract:
The system prompts the Administrator to enter the new duration and validates
the renewal before updating the contract
Print Contract:
The system generates a PDF or printable version of the contract, ready for
download or printing
2.1.4.5 Detailed use case diagram

Figure 2.13 Teacher agent usecase

Hoang Ha – Class 20SE6 26

Building a management system for a chain of private training and student care centers using AI technology

Figure 2.14 Central management agent usecase

Hoang Ha – Class 20SE6 27

Building a management system for a chain of private training and student care centers using AI technology

Figure 2.15 Parent agent usecase

Figure 2.16 Student management function use case

Hoang Ha – Class 20SE6 28

Building a management system for a chain of private training and student care centers using AI technology

Figure 2.17 Usecase of classroom management function

2.1.4.6 Detailed Use Case Specification

Use-case Name Student Management

Actor Admin, Center Management, Teacher, Staff

Description Manage student information, including creating, updating,

deleting, viewing, and searching for records. Additional
functions include emailing parents.

Pre-Condition User has logged in successfully

Post-Condition Student data is managed or updated as required

Alternative Flows 1. Create new: Add a new student record.

2. Delete: Remove an existing student record.
3. Update: Modify student details.
4. Search: Find student information.
5. View information: Display student details.
6. Email parents: Send communication to parents.
Table 2.1 Student management function use case specification

Use-case Name Student Attendance Check-In

Actor Student Attendance

Hoang Ha – Class 20SE6 29

Building a management system for a chain of private training and student care centers using AI technology

Description Helping students to check attendance through facial

recognition.
Pre-Condition Camera ready, student face prepared for scanning
Post-Condition Attendance recorded successfully.
Alternative Flows 1. Face detection extracts the image from the camera
2. The system processes and recognizes the face
3. Save the attendance information into the database
4. If the system cannot recognize the face → Request a
retry
Table 2.2 Student attendance use case specification

Use-case Name Class Management

Actor Administrator, Center Management, Teacher, Staff
Description Manage class information, including creating, updating,
deleting, viewing details, and logging into the system.
Pre-Condition User has successfully logged into the system
Post-Condition Class data is successfully updated or managed
Table 2.3 Classroom management use case specification
2.1.5 Class diagram
2.1.5.1 Identify the classes
- Center class
- Class class
- Block class
- Semester class
- Parent class
- Student class
- Subject class
- Schedule class
- Credit class

Hoang Ha – Class 20SE6 30

Building a management system for a chain of private training and student care centers using AI technology

- Learning outcome class

- Student health class
- Diet class
- Food class
- Admission class
- Contract class
- Personnel class
- Extracurricular activity class
2.1.5.2 Analysis class diagram

Figure 2.18 Class diagram

Hoang Ha – Class 20SE6 31

Building a management system for a chain of private training and student care centers using AI technology

2.1.6 Activity chart

Figure 2.19 Login Activity Diagram

Hoang Ha – Class 20SE6 32

Building a management system for a chain of private training and student care centers using AI technology

Figure 2.20 Activity diagram for adding new teachers

Hoang Ha – Class 20SE6 33

Building a management system for a chain of private training and student care centers using AI technology

Figure 2.21 Roll call activity diagram

Hoang Ha – Class 20SE6 34

Building a management system for a chain of private training and student care centers using AI technology

Figure 2.22 Chatbot Activity diagram

Hoang Ha – Class 20SE6 35

Building a management system for a chain of private training and student care centers using AI technology

2.1.7 Sequence diagram

Figure 2.23 Login Sequence Diagram

Figure 2.24 Sequence diagram adding new data

Hoang Ha – Class 20SE6 36

Building a management system for a chain of private training and student care centers using AI technology

2.2. DATABASE
The database is a very important part of the private student care and training center
chain management system, it stores and manages relevant information. Because the
number of tables is too large, a brief presentation of the main tables such as: Account,
center, classroom, employee, enrollment, parents, student, subjects, study_result,...
Table: account
No Logical name Data type Default Primary Remarks Index
key

1 id bigint N/A o Unique Clustered

identifier
2 account varchar(255) N/A User
account
name
3 password varchar(255) N/A User
password
(hashed)
4 email varchar(255) N/A User email
address
5 status tinyint(1) 1 Account
status
(active/inact
ive)
6 date_created timestamp current Date the
_timest account was
amp() created
7 last_login datetime NULL Last login
timestamp
8 password_salt varchar(255) NULL Salt for
password
hashing
9 account_type bigint NULL Foreign key
to account
type

Hoang Ha – Class 20SE6 37

Building a management system for a chain of private training and student care centers using AI technology

10 google_id varchar(255) NULL Google

account ID
(if any)
11 created_at timestamp NULL Record
creation
timestamp
12 updated_at timestamp NULL Last update
timestamp
Table 2.4 Account table description
Table: center
No Logical name Data type Default Primar Remarks Index
y key

1 id bigint N/A o Unique Clustered

identifier
2 name varchar(255) N/A Center name
3 email varchar(255) N/A Center email
4 phone varchar(255) NULL Center phone
number
5 address text NULL Center
address
6 date_created timestamp current Date the
_timest center was
amp() created
7 desc text NULL Description
of the center
8 account_atm varchar(255) NULL ATM
account
number
9 employee_id bigint NULL Foreign key
to employee
10 status int(11) NULL Center status
11 created_at timestamp NULL Record
creation

Hoang Ha – Class 20SE6 38

Building a management system for a chain of private training and student care centers using AI technology

timestamp
12 updated_at timestamp NULL Last update
timestamp
Table 2.5 Center table description
Table: classroom
No Logical name Data type Default Primar Remarks Index
y key

1 id bigint N/A o Unique Clustered

identifier
2 class_name varchar(255) N/A Name of the
class
3 class_code varchar(255) N/A Code Indexed
identifying
the class
4 class_size int(11) N/A Number of
students in
the class
5 center_id bigint NULL Foreign key Indexed
to center
6 description text NULL Description
of the class
7 employee_id bigint NULL Foreign key Indexed
to employee
8 class_block_id bigint NULL Foreign key Indexed
to class
block
9 created_at timestamp NULL Record
creation
timestamp
10 updated_at timestamp NULL Last update
timestamp
Table 2.6 Classroom table description

Hoang Ha – Class 20SE6 39

Building a management system for a chain of private training and student care centers using AI technology

Figure 2.25 Database

Hoang Ha – Class 20SE6 40

Building a management system for a chain of private training and student care centers using AI technology

2.3. HOW RESTFUL APIS HANDLE DATA

RESTful API can be used to send and receive data from the server and integrate it
into the central chain management system. By using the POST method, RESTful API
can retrieve related exercises, student lists, and more. When users access specific
screens, the system utilizes the Http library to establish a connection with the RESTful
API, facilitating data retrieval from the server and displaying it on the screen.

Figure 2.26 List API

Figure 2.27 API return result

- The functionality of viewing related exercises through the POST method of
RESTful API is designed to enable users to access additional exercises with similar
content, aiding students in enhancing their knowledge. Data is sent to FastAPI or Flask

Hoang Ha – Class 20SE6 41

Building a management system for a chain of private training and student care centers using AI technology

hosted on an AWS server via the POST method, and the results are returned for further
processing and display.
Parameters used in the API to view suggested exercises can include:
+ exercise: Contents of the current exercise
2.4. FACE DETECTION AND RECOGNITION
2.4.1 Overview of face detection and recognition
2.4.1.1 Concept of face detection and recognition
Face detection is a term in the field of computer vision and image processing,
referring to the process of recognizing faces appearing in images or videos through
machine learning methods.
In parallel, face recognition has a different definition than face detection. It is the
process of verifying and identifying a particular face by comparing the student's facial
features with the data available in the database. Face recognition is not only an
important research area in computer vision but also a specialized branch of research in
human biometrics.
2.4.1.2 Impact of face detection and recognition methods on attendance
systems
Face detection and recognition methods, especially Deep Learning-based
methods, play an important role in attendance systems in education, bringing many
benefits to schools and educational institutions. Below are the main benefits of
applying these techniques:
Helps reduce attendance time, increase accuracy and efficiency in student
management.
Improve user experience, increase transparency and security in tracking and
verifying identity.
Enhance the effectiveness of student engagement management and tracking.
Contribute to improving the quality of teaching and management in educational
institutions, helping to save time and resources.
Deep learning facial recognition methods not only help optimize the attendance
process but also create a more efficient and safer learning environment.

Hoang Ha – Class 20SE6 42

Building a management system for a chain of private training and student care centers using AI technology

2.4.2 Applying face detection and recognition to the system

Face Detection and Recognition Type: The attendance system uses AI to recognize
faces through images, applying the Deep Learning method, to bring high accuracy in
student authentication. The system collects data from videos or images of students'
faces, along with personal information such as full name, class and student ID number,
to proceed with labeling. From the collected video or facial image data, the system will
use advanced face recognition algorithms such as MTCNN and FaceNet to analyze the
face and extract images from the video. This information will be matched with the
existing database to authenticate and accurately take attendance of students.
2.4.2.1 Concepts of face recognition in MTCNN and FaceNet models
MTCNN (Multi-task Cascaded Convolutional Networks):
*MTCNN model (Multi-task Cascaded Convolutional Networks)
MTCNN is a popular and effective deep network for face detection and facial
landmark prediction. The network consists of 3 processing stages through 3 sub-
networks:
P-Net (Proposal Network)
Fast detection of potential face regions using sliding window mechanism on multi-
scale input images

Figure 2.28 Proposal Network

The input image is then divided into "patches" (small regions), and each region
is fed into P-Net, a small CNN network with a simple architecture for analysis.

Hoang Ha – Class 20SE6 43

Building a management system for a chain of private training and student care centers using AI technology

Figure 2.29 P-Net CNN network

Returns bounding boxes and confidence scores
Apply Non-Maximum Suppression (NMS) algorithm to remove overlapping
regions, keeping the boxes with the highest scores.
R-Net

Figure 2.30 R-Net

Refine and filter the bounding boxes from P-Net, focusing on specific regions
instead of scanning the entire image.
Perform:
(1) Remove invalid boxes.
(2) Refine the size and position of bounding boxes.
Continue using NMS to eliminate overlap, improve output quality.
O-Net

Figure 2.31 O-Nets

Hoang Ha – Class 20SE6 44

Building a management system for a chain of private training and student care centers using AI technology

Accurately locate faces and predict landmarks (eyes, nose, mouth)

Fine-tune bounding boxes and confidence scores with the highest level of detail
Apply NMS to filter out redundant boxes
Final result: Exact bounding boxes, confidence scores and landmarks.
*FaceNet Model
Similar Machine Learning Methods in Face Authentication
One-shot Learning
Machine learning methods help identify or classify new objects with very little
data.
Instead of direct classification, the model learns the similarity between samples
by mapping them to vector embeddings in the feature space and comparing the
distance between the vectors (Euclidean or cosine)

Figure 2.32 Siamese Networks Architecture

Siamese Networks
Triplet Networks
Triplet Networks are an extended variant of Siamese Networks, in which three
input patterns are used to learn to distinguish classes from each other. These three
patterns include:
 Anchor: Sample of facility to be inspected.
 Positive: Same class as anchor.

Hoang Ha – Class 20SE6 45

Building a management system for a chain of private training and student care centers using AI technology

 Negative: Different class pattern with anchor.

The goal of Triplet Networks is to optimize so that the distance between the
Figure 2.33 Siamese Networks Architecture
anchor and the positive is smaller than the distance between the anchor and the
negative, with a margin of safety. This is done through the triplet loss function, which
ensures that the classes in the feature space are well separated.
Prototypical Networks
Prototypical Networks is a more modern approach that simplifies the
classification process in feature space. The main idea is to use the average (prototype)
of the embedding vectors in the feature space to represent each class. For each class,
the system calculates the average vector of the examples in that class (prototype).
When a new example appears, the system calculates the distance from the vector of
that example to the existing prototypes and determines the closest class.
Facenet

Figure 2.34 FaceNet Algorithm

Hoang Ha – Class 20SE6 46

Building a management system for a chain of private training and student care centers using AI technology

FaceNet is a powerful and efficient algorithm designed to solve face recognition

and face verification problems. It does not focus on directly classifying faces into
categories (as is traditional), but rather learns to represent faces as vector embeddings
in a feature space, where the distance between embeddings reflects how similar the
faces are. This places FaceNet in the context of One-shot Learning and Learning
Similarity techniques.
Instead of classifying faces based on class labels, FaceNet learns a mapping
function from the image space to the embedding space such that:
- Faces of the same person are mapped close together in the embedding space.
- Faces of different people are mapped far apart.
Distances in the embedding space (usually Euclidean or cosine distance) are
used to compare and evaluate the similarity between two faces. This is very suitable
for One-shot Learning methods, where only a few samples (or even one sample) are
needed to determine or authenticate the identity of a person.
FaceNet is built on a deep neural network architecture such as Inception-Resnet
or MobileNet ....., and the idea of optimizing the distance in the embedding space is
implemented through the triplet loss function:

Hoang Ha – Class 20SE6 47

Building a management system for a chain of private training and student care centers using AI technology

Triplet Loss
Each time of training, the network receives three images:
Anchor: Original image of a person.
Positive: Another picture of the same person as anchor.
Negative: Image of another person.
The loss function is optimized to:
The distance between the anchor and the positive (same person) is smaller than
the distance between the anchor and the negative (different people), with a margin of
safety).
Loss function formula:

Figure 2.35 Triplet loss function

Figure 2.36 Loss function formula

In there:
- d(A, P): Distance between anchor and positive.
- d(A, N): Distance between anchor and negative.
- margin: A non-negative value, ensuring that the distance between layers is large
enough.
Meaning of triplet loss:
- Reduce the distance between embeddings of the same person, helping to group
the person's features closer together.
- Increase the distance between embeddings of different people, ensuring there is
no overlap between layers in the feature space.

Hoang Ha – Class 20SE6 48

Building a management system for a chain of private training and student care centers using AI technology

- Transform the embedding space into a metric space, where distance reflects
similarity
2.4.2.2 Results of applying face detection and recognition method to the
system
Compare the results of 2 facial recognition methods:
ModelTable 2.7 Compare
Number ofthe results of 2 facial recognition
Accuracy CPU runtimemethods
Number of
samples (s) parameters

VGG19 + 2,606 90% 1.37 143M

Softmax
Resnet50 + 2,606 95.2% 1.72 25M
Triplet

After applying the face detection and recognition method to the system, the
detection and recognition results achieved certain expectations.
The automatic attendance results in the database are returned including attendance
code, student code, attendance status

Figure 2.37 Face recognition results

This result not only helps reduce the workload for teachers but also improves
the quality of classroom management, connecting schools, teachers and students.
2.5. APPLYING THE K-NEAREST NEIGHBORS MODEL TO ANALYZE
AND EVALUATE STUDENT BEHAVIOR
2.5.1 K-Nearest Neighbors Overview
K-Nearest Neighbors (KNN) is a machine learning algorithm belonging to the
Supervised Learning group, outstanding with its simplicity and flexible application in
many fields, especially classification problems. (Classification) and regression

Hoang Ha – Class 20SE6 49

Building a management system for a chain of private training and student care centers using AI technology

(Regression). The algorithm works based on the principle: "New data points will be
predicted based on the labels or values of the k nearest neighboring data points in the
feature space."
KNN does not require a model training process like other algorithms but directly
uses all training data for prediction. When a new data point becomes available, the
algorithm calculates the distance between this point and all the data points in the
training set, often using methods such as Euclidean, Manhattan, or Minkowski
distance. Then, the algorithm determines the k nearest neighbors
2.5.2 Overview of the Student Stress Factors Dataset: A Comprehensive
Analysis

Figure 2.38 Student Stress Factors Dataset

The "Student Stress Factors: A Comprehensive Analysis" dataset is designed to
analyze factors that affect students' stress levels, focusing on aspects such as health,
psychology, society, and living environment. The dataset includes many features such
as sleep quality, academic pressure, teacher-student relationships, personal safety, and
other social factors such as support from family, friends, or bullying. The values in the
data are often rated on a scale of 1 to 5, which helps standardize and make it easy to
process. With the main goal of supporting research on the causes of stress and
predicting students' stress levels through machine learning algorithms such as K-
Nearest Neighbors (KNN) or Logistic Regression, this dataset also aims to propose
solutions to improve the quality of life and learning environment. However, the dataset
also has limitations such as possibly containing noise or missing data, which need to
be processed before use. With its diversity and comprehensiveness, this dataset brings

Hoang Ha – Class 20SE6 50

Building a management system for a chain of private training and student care centers using AI technology

great value for research and practical applications in improving students' mental health
and academic performance.
2.5.3 Make predictions on machine learning models
Use other algorithms like Random Forest and Support Vector Machine to make
predictions about the dataset
K-Nearest Neighbors Random Forest Support Vector Machine
86.36% 85.00% 86.16%
Table 2.8 Comparing the prediction accuracy of 3 machine learning models

Figure 2.39 Prediction results of KNN model

2.6. APPLYING CONTENT-BASED RECOMMENDATION SYSTEMS TO
THE SYSTEM
2.6.1 Overview of Content-based Recommendation Systems
Content-based Recommendation Systems is a recommendation method based on
analyzing the features of objects (content, attributes) to find similar objects and
recommend them to users. This system uses the target's descriptive information (e.g.
text, keywords, product attributes) and builds a model based on the relationship
between these features.
In the field of natural language processing (NLP), Content-based Systems often use
techniques such as TF-IDF, Word Embeddings, or Sentence Embeddings to represent
text into numeric vectors and calculate the similarity between objects.

Hoang Ha – Class 20SE6 51

Building a management system for a chain of private training and student care centers using AI technology

Figure 2.40 Content-based Recommendation Systems

For example:
If a user searches for an exercise on "multiplying fractions," the system will return
similar exercises based on the content and semantics of the query.
2.6.2 Overview of the RAG (Retrieval-Augmented Generation) method
Retrieval-Augmented Generation (RAG) is an advanced method in natural
language processing (NLP), combining information retrieval (retrieval) and text
generation (generation).
The goal of RAG is to improve the quality and accuracy of large language
models (LLMs) by allowing the model to retrieve external information from real-
world data sources during text generation. This helps the model rely on the latest
knowledge, instead of just relying on pre-trained data.

Hoang Ha – Class 20SE6 52

Building a management system for a chain of private training and student care centers using AI technology

Figure 2.41 RAG method

2.6.3 Working principle
Content-based recommendation systems work in the following steps:
Step 1: In case of text, common techniques are:
TF-IDF (Term Frequency-Inverse Document Frequency): Convert text to
vector matrix based on word frequency
Word Embedding: Use models like Word2Vec, GloVe to generate word vectors
Sentence Embedding: Use models like SBERT to create vectors that represent the
entire text
Step 2: Similarity Computation
Measures the similarity between target feature vectors. Popular methods:
+ Cosine Similarity: Measure the angle between two vectors in vector space

+ Euclidean Distance: Distance between two vectors

Items with high similarity will be recommended
Step 3: Ranking and returning results

Hoang Ha – Class 20SE6 53

Building a management system for a chain of private training and student care centers using AI technology

The items are sorted by similarity and returned to the user

2.6.4 Overview of the keepitreal/vietnamese-sbert model
The keepitreal/vietnamese-sbert model is a version of Sentence-BERT (SBERT)
specially trained for Vietnamese. SBERT is a variant of the BERT (Bidirectional
Encoder Representations from Transformers) model, which is fine-tuned to generate
semantic representation vectors of sentences. This model helps to calculate semantic
similarity between sentences or text paragraphs efficiently.
2.6.5 System architecture

Figure 2.42 Recommendation Systems workflow

The exercise recommendation system operates in a sequential process, starting
with the user sending a request via the Flask API. Flask acts as a communication
gateway between the user and the system, receiving and processing requests.

After receiving the request, the system will load exercise data from the CSV file
via CSVLoader. This is the data preparation step, ensuring that the system has all the
necessary exercise information to proceed with analysis and processing.

Next, the exercise data is converted into embedding vectors using the
Vietnamese-SBERT model. SBERT is an upgraded version of BERT, optimized for

Hoang Ha – Class 20SE6 54

Building a management system for a chain of private training and student care centers using AI technology

comparing the similarity between sentences, helping to represent text content in vector
form, thereby supporting more effective search for related exercises.

These vectors are then stored in ChromaDB, a vector database capable of

quickly searching and retrieving similar exercises. When a user submits a query, the
system searches ChromaDB and returns exercises that are similar or most relevant to
the user's request.

The entire process ensures that users receive quick and accurate suggested
exercises, improving the learning experience and making content searching more
efficient.

2.7 APPLY THE LLM MODEL TO BUILD CHATBOTS IN THE SYSTEM

2.7.1 Overview of chatbots
Chatbot is software that simulates conversations with humans via text or voice,
helping to automatically answer questions, provide customer support, and process
information. There are two main types: rule-based chatbots, which operate according
to existing scenarios, and AI chatbots that use artificial intelligence to respond more
intelligently. Chatbots bring great benefits such as saving costs, operating 24/7, and
improving customer experience. They are widely used in customer service, healthcare,
finance, and e-commerce, with prominent platforms such as ChatGPT and Google
Bard.
2.7.2 Overview of the LLM model
Large Language Model (LLM) is a type of artificial intelligence (AI) model trained
on huge volumes of text data, capable of understanding, processing and generating
natural language. LLM is gradually becoming the core of AI applications in many
fields such as customer care, healthcare, education, programming and entertainment.
Operational process of the LLM model:
Step 1: Collect data
Collect text data from books, newspapers, blogs, forums, source code, scientific
documents, websites,...
The more diverse the data, the better the model is able to process natural language
Step 2: Preprocess data
Eliminate irrelevant or noisy data

Hoang Ha – Class 20SE6 55

Building a management system for a chain of private training and student care centers using AI technology

Split text into tokens (words or phrases) for easy processing by the model
Step 3: Training on Transformer architecture
Transformer is the core architecture of LLM (introduced by Google in 2017)
Based on Attention Mechanism, the model learns to determine the focus and main
meaning of a sentence
The model learns by predicting the next word in a sentence based on the surrounding
context

Figure 2.43 Transformer Architecture

2.7.3 Operational process of chatbot in the system
The Qwen2.5-1.5B-Instruct model is an LLM developed by Alibaba, with 1.5
billion parameters, optimal for instruction-tuning tasks. This helps the model
understand user commands and provide more accurate and flexible responses than
traditional models.
Qwen2.5-1.5B-Instruct is the ideal choice for chatbot or virtual assistant systems
that need a balance between performance, cost and rapid deployment. With the
adjustment of structural guidelines, this model provides a natural, smooth and precise
user experience, suitable for this topic.

Hoang Ha – Class 20SE6 56

Building a management system for a chain of private training and student care centers using AI technology

Figure 2.44 Chatbot Activity Flow

This chatbot system works by receiving questions from users and processing
them through two main directions to find the appropriate answer. When the user
submits a question, the system will analyze and select a processing method based on
the available data source.
If the required information is in the database, the system will send a request to
OpenAI to generate an SQL statement based on the table structure and available
information. The query results are then returned to the user in the form of an answer.
In the case of data in text files (txt files), the system uses the Qwen/Qwen2.5-
1.5B-Instruct model to convert the content into vectors, search for similar content, and
provide appropriate answers.
The entire process ensures that the system can answer questions accurately, flexibly,
and quickly, taking advantage of both data in the database and text files to bring the
best experience to users.

Hoang Ha – Class 20SE6 57

Building a management system for a chain of private training and student care centers using AI technology

2.8 DEPLOY APIS TO AWS

Figure 2.45 Deploy APIs to AWS

I use Docker to deploy FastAPI applications on Amazon Web Services (AWS).
By containerizing FastAPI, the application becomes portable and easily scalable,
enabling smooth deployments across different environments. Docker images are built
locally and pushed to AWS services like Elastic Beanstalk, ECS, or EC2, facilitating
automated deployments and minimizing manual intervention.
This method leverages Docker’s isolation and orchestration features to ensure
that the FastAPI application runs consistently regardless of the environment. By using
AWS, the infrastructure is highly available and can handle large volumes of traffic,
ensuring low latency and high performance. Additionally, Docker allows for rapid
updates and rollback capabilities, providing flexibility during development and
deployment cycles. This approach enhances the overall reliability of the system and
simplifies scaling by spinning up multiple containers when traffic increases, ensuring a
seamless experience for users.
2.9 END OF CHAPTER 2
After this chapter, we have mastered the system analysis and design process of
a chain of centers management system and deeply understood the application of AI.
The FaceNet model supports facial recognition for automatic attendance, KNN
predicts students' stress levels, while RAG, vietnamese-sbert and Chroma suggest
appropriate exercises. Chatbot integrated with LLM helps answer questions, creating

Hoang Ha – Class 20SE6 58

Building a management system for a chain of private training and student care centers using AI technology

a solid foundation for building a modern central chain management system in chapter
3.

Hoang Ha – Class 20SE6 59

Building a management system for a chain of private training and student care centers using AI technology

CHAPTER 3. IMPLEMENTATION OF BUILDING SYSTEM

3.1 SYSTEM OVERVIEW

Languages used:
- About website:
+ Regarding the interface, we use the learned languages such as HTML, CSS,
JavaScript, and also use Bootstrap4 and jquery libraries to build a website interface.
+ Regarding system functions, the main language used is PHP and the Laravel
11.x Framework is also used.
- About AI:
+ Application of FaceNet model in face recognition
+ Applying KNN algorithm to predict students' stress level based on data from
StressLevelDataset.csv and survey information.
+ Using the RAG method and the keepitreal/vietnamese-sbert and Chroma
models to search and suggest similar exercises
+ Use the LLM (Large Language Model) model to implement the chatbot's
question answering function
- MySQL database management system
- The system uses RESTful API to get data from the server.
3.2 RESULTS ACHIEVED
3.2.1 About the management system for administrators and employees
Administrators can manage the roles of personnel in the system.

Hoang Ha – Class 20SE6 60

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.46 Role Management

Administrators can manage staff and parent accounts in the system. After
creating a staff or parent, the system will automatically create an account for that user
with a default password.

Figure 3.47 Account Management

Administrators can manage the current class blocks available in the system.

Hoang Ha – Class 20SE6 61

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.48 Class block management

The administrator can manage the semesters of the system. This is to determine
which semester has a study period from which date to which date.

Figure 3.49 Semester Management

Administrators can manage centers that have signed up for a contract to use the
system.

Hoang Ha – Class 20SE6 62

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.50 Center Management

For the central role, the center manager can manage information of the staff and
teachers in the center.

Figure 3.51 Human resource managementss

Center managers or staff can manage the current enrollment status of the center.
Can view and update the status of enrollment activities and can evaluate the
enrollment level at the center.

Hoang Ha – Class 20SE6 63

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.52 Admissions Management

The center manager or staff can manage the contract after signing with the
parents to let the student study at the center. The staff can print the contract and store
the contract information on the system.

Figure 3.53 Contract Management

Center administrators or staff can manage parent information. Parent contact

information can be viewed when needed.

Hoang Ha – Class 20SE6 64

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.54 Parental Management

Center administrators or staff can manage information of current students

studying at the center.

Figure 3.55 Student Management

Center managers or staff can contact parents via email. After composing the
email content, the system will automatically send the email to all parents in the class.

Hoang Ha – Class 20SE6 65

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.56 Contact parents

Figure 3.57 Email notification system

Center administrators or staff can manage classes. Can view the list of students
in that class and the courses that class is studying.

Hoang Ha – Class 20SE6 66

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.58 Class Management

Center administrators or staff can manage subjects

Figure 3.59 Subject Management

Center managers or staff can manage additional reference exercises, which will
be displayed on the parent interface to help students review what they have learned.

Hoang Ha – Class 20SE6 67

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.60 Assignment Management

Center managers or staff can manage the module classes to arrange and schedule each
class in accordance with the criteria of education.

Figure 3.61 Course Management

Center managers, staff or teachers can manage students' learning results on the
system, and after summarizing, the results will be sent to parents.

Hoang Ha – Class 20SE6 68

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.62 Managing learning outcomes

Center managers and medical staff can manage students' health examination
results. After exporting the sample file on the system and entering the examination
results, the system will save the results to the database.

Figure 3.63 Student Health Management – 1

Hoang Ha – Class 20SE6 69

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.64 Student Health Management – 2

Figure 3.65 Student Health Management – 3

Center managers and staff can manage extracurricular activities organized by

the center to create a playground for students after school hours.

Hoang Ha – Class 20SE6 70

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.66 Extracurricular Activities Management

Center managers and staff can manage student meals. After scheduling weekly
meals, at each meal, the kitchen staff will update the image of the day's dishes on the
system. This helps parents track their children's meals on the system to improve the
quality of the center.

Figure 3.67 Food Portion Management – 1

Hoang Ha – Class 20SE6 71

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.68 Food Portion Management – 2

Centers can manage their own food to ensure food safety and hygiene for
students.

Figure 3.69 Food Management

Centers can set current semester and tuition fees for semesters at the center

Hoang Ha – Class 20SE6 72

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.70 Set up semester

Figure 3.71 Set up tuition

System login interface, users can log in with gmail to the system without having
to enter a password or account.

Hoang Ha – Class 20SE6 73

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.72 Login to the system

When students arrive at school, the staff or interface will press the attendance
button for the students, when the students show their faces to the camera, the system
will automatically take attendance. This is applied with AI technology to recognize
faces in the system to help reduce the workload for staff.

Figure 3.73 Attendance screen

Hoang Ha – Class 20SE6 74

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.74 Track attendance

In addition, the system also has functions such as statistics.

Figure 3.75 General statistics

Hoang Ha – Class 20SE6 75

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.76 Revenue statistics

Figure 3.77 Stress level survey statistics

3.2.2 About the interface for parents and new users

When there is an extracurricular activity that is suitable for the parent's child,
the system will display that advertisement on the home screen for 10 seconds so that
the parent can get the information.

Hoang Ha – Class 20SE6 76

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.78 Home

After parents fill out the form to submit if they have a need for enrollment, the
system will push a notification to the manager or staff of that center so that the
consultant can contact the parents as soon as possible.

Figure 3.79 Admissions Consulting

When parents visit this page, they can track student information, such as
academic performance, tuition, health, and registered extracurricular activities.

Hoang Ha – Class 20SE6 77

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.80 Student Information – 1

Figure 3.81 Student Information – 2

Hoang Ha – Class 20SE6 78

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.82 Student Information – 3

Figure 3.83 Student Information – 4

Once users have tracked their child's tuition status for the semester, they can
pay tuition online via the Momo payment gateway without having to go to the center.

Hoang Ha – Class 20SE6 79

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.84 Pay tuition fees via Momo portal

Parents can monitor the subjects that students study at the center.

Figure 3.85 List of subjectss

Hoang Ha – Class 20SE6 80

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.86 Course details

Parents can view information about the subject's instructor, course materials,
and exercises to further reinforce students' knowledge.

After parents choose to do an exercise, they will be taken to the exercise details,
including: the assignment, instructions on how to do it, parents can enter the answer
after the student has done it or view the answer to that exercise.

Figure 3.87 Exercise details

When parents want their children to do exercises similar to that exercise for
more practice, parents click the suggested exercises button, the system will rely on AI

Hoang Ha – Class 20SE6 81

Building a management system for a chain of private training and student care centers using AI technology

to get exercises with similar content to that exercise and display the 5 most similar
exercises.

Figure 3.88 Suggested exercises

In addition, if parents do not understand how to do exercises such as essays, the

system will integrate a chatbot to provide suggestions and comments to solve the
exercise in the simplest way.

Figure 3.89 Instructions and comments on exercises

Currently, the eating situation of students at the center is a matter of public

concern. With this system, parents can monitor the eating plan of students at the
center, can see what their children eat that day and the center will update the image of
that meal for parents' peace of mind.

Hoang Ha – Class 20SE6 82

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.90 Serving Size

Figure 3.91 Meal details

When parents see that their children are showing signs of stress while studying
at the center, parents can survey students with a few questions on the system screen,
from which the system will apply AI algorithms to predict the stress and pressure of
students through a data set in order to promptly have a solution if the student is having
that problem.

Hoang Ha – Class 20SE6 83

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.92 Study Pressure Survey – 1

Based on this prediction, the system will notify parents about the level of stress
and pressure of students. The system will integrate chatbots to give advice to parents
when their children are experiencing stress.

Figure 3.93 Study Pressure Survey – 2

Parents can monitor the schedule at the student's center.

Hoang Ha – Class 20SE6 84

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.94 Class schedule

In addition to studying, parents can register their children to participate in

extracurricular activities organized by the centers, to help students relax after stressful
school hours and enhance solidarity.

Figure 3.95 Extracurricular activities

Hoang Ha – Class 20SE6 85

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.96 Register for extracurricular activitiess

In addition, the system also applies AI technology to build a chatbot to support

parents and users when they want to answer questions about the center or the center's
services to help users have the best experience.

Figure 3.97 Chatbot interface

In addition, the system also has other pages such as: login, about us, contact,...

Hoang Ha – Class 20SE6 86

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.98 Login page

Figure 3.99 About Us Page

Hoang Ha – Class 20SE6 87

Building a management system for a chain of private training and student care centers using AI technology

Figure 3.100 Contact page

3.3 SUMMARY
This chapter presents the process of building a management system for a chain
of private training and student care centers using AI. The website interface is designed
using HTML, CSS, JavaScript combined with Bootstrap4 and jQuery, while the main
functions are developed using PHP and Laravel 11.x. The system uses the FaceNet
model for facial recognition, KNN to predict stress levels from survey data, and RAG
with keepitreal/vietnamese-sbert and Chroma models to search for exercises. Chatbot
integrated with LLM supports answering questions. Data is managed using MySQL
and synchronized via RESTful API, ensuring the stability, security and scalability of
system.

Hoang Ha – Class 20SE6 88

Building a management system for a chain of private training and student care centers using AI technology

CONCLUSIONS AND SUGGESTIONS

This research focuses on building a chain management system for private student
training and care centers, integrating AI technology to improve management efficiency
and user experience. The application of technologies such as facial recognition, stress
monitoring and intelligent response chatbots opens up new directions in ensuring
safety, improving education quality and supporting center management. The research
results show contributions to the management of inclusive central chains:
- Enhance security and automatic attendance: Face recognition system helps
quickly authenticate, automatically take student attendance and prevent
unauthorized intrusion at centers, ensuring safety and accuracy in
administration. student psychology.
- Monitor student stress: The system monitors and evaluates stress levels
through surveys, promptly warns and provides appropriate solutions, ensuring
students are supported and rested when needed.
- AI suggests related exercises: The AI system automatically analyzes and
suggests exercises appropriate to the student's level and status, contributing to
personalizing learning and improving teaching effectiveness.
- Smart feedback via chatbot: Integrated AI chatbot provides quick and
accurate information about class schedules, student status and frequently asked
questions, effectively supporting teachers, students and parents.
- Building a safe and modern learning environment: Integrating AI
technology helps create a smart learning environment, improve educational
quality and promote students' proactive learning.
The research has brought new insight and contributed to promoting innovation in
private center management, especially in the context of the increasing need to use
smart and comprehensive management tools.

Hoang Ha – Class 20SE6 89

 REFERENCES

[1] Lê Đăng Hiển, Lê Văn Nam (2020), “Ý định sử dụng trí tuệ nhân tạo trong
học tập của sinh viên: Tổng quan nghiên cứu và khung phân tích”,

http://thuvienlamdong.org.vn:81/bitstream/DL_134679/23263/1/CVv139S2
42020065.pdf.
[2] Khoa học và công nghệ (2021), “NHẬN DIỆN KHUÔN MẶT SỬ DỤNG
MẠNG NƠRON TÍCH CHẬP XẾP CHỒNG VÀ MÔ HÌNH FACENET”,
https://jst-haui.vn/media/30/uffile-upload-no-title30548.pdf.
[3] Bill Lubanovic (2023), “FastAPI Modern Python Web Development”.
[4] Harshit Kumar Chaubey; Gaurav Tripathi; Rajnish Ranjan; Srinivasa k.
Gopalaiyengar (2024), “Comparative Analysis of RAG, Fine-Tuning, and
Prompt Engineering in Chatbot Development”,
https://ieeexplore.ieee.org/abstract/document/10691338.
[5] Lishan Wang (2019), “Research and Implementation of Machine Learning
Classifier Based on KNN”.
[6] Sumit Kumar Dam, Choong Seon Hong, Yu Qiao, Chaoning Zhang (2024),
”A Complete Survey on LLM-based AI Chatbots”,
https://arxiv.org/abs/2406.16937.
[7] Fuzheng Zhang, Nicholas Jing Yuan, Defu Lian, Xing Xie, Wei-Ying Ma
(2016), “Collaborative Knowledge Base Embedding for Recommender
Systems” ,
https://dl.acm.org/doi/abs/10.1145/2939672.2939673.
[8] Lufeng Han (2018), “Analysis of New Advances in the Application of
Artificial Intelligence to Education”,
https://www.atlantis-press.com/proceedings/iceemt-18/55907823.

vii