Vol. 9, No.

2 | July – Dec 2025

AI-Driven Recruitment and Hiring Automation through

CV Scanning and AI Agents
Nareta Kataria1 , Dr. Khalid Rasheed1
Abstract:
The presented work is dedicated to the development of an AI agent that will perform automated CV sorting
and evaluation to assist current hiring processes with a supervised machine learning model. A Random
Forest Regressor containing 300 trees were trained on structured features based on the unstructured CV
information including years of experience, number of related skills, education level, overlap of skills with
the job description, and coverage ratio. The 50 actual CVs dataset was divided to 80 percent training set
and 20 percent testing set and the ground truth target used in supervised learning was HR inspired rule
based scoring. The model realized an impressive predictive power, as seen with the R² of 0.94 and Mean
Absolute Error of 4.46 which signifies a significant agreement with human based ratings. The trained model
is integrated into an AI agent that automates the ranking of the candidates, short listing and scheduling of
interviews. The system is particularly effective at unloading HR staff and increasing the efficiency of the
recruitment process; nevertheless, some of the key issues are the bias in the training data, the tendency of
the system to treat different candidates with unequal consideration, and the limited interpretability of the
model. The study highlights the importance of introducing feedback loops, reducing bias, and establishing
transparency to ensure that AI-based recruitment systems were ethically, reliably, and humanly deployed.

Keywords: Natural Language Processing (NLP), Artificial intelligence (AI), Autonomous Recruitment
Systems, Predictive Analytics.

1. Introduction sorting and assessment tools based on

artificial intelligence (AI) can analyze
A. Background unstructured data, quantify the qualification,
Every organization should have a experience, and suitability of a candidate to a
recruitment section, yet the traditional hiring job, and generate first level scores that can be
processes are often tedious, inaccurate and related to evaluation criteria. The aim of such
prejudiced. The HR specialists need to go systems is to reduce bias, accelerate the hiring
through a huge number of resumes manually, process, and reduce the workload of the HR
screen the qualifications of the applicants, and personnel by emulating human decision
form snap judgments, which may result in making. In spite of these advantages, AI-based
errors, delays, and discrimination. The need is hiring has some disadvantages. Biases can be
to have automated, effective and urgent. fair passed on in training data, model decisions
procedures within the recruitment field as a cannot be transparent, and it is necessary to be
result of the growing complexity and numbers cautious to ensure fairness across a. range of
of applications. Artificial intelligence (AI) has candidate profiles. These problems should be
massive potential that can transform the addressed by designing models carefully,
recruitment sector by automating repetitive continuing to improve and add feedback
processes, standardizing evaluation processes, systems that improve accuracy, equity and
and enhancing the accuracy of decisions. CV interpretability. This paper introduces the

1
Department of Computer Science Szabist University Karachi
[Email]: mscs2473113@[Link] , khalid.shaikh82@[Link]
SJCMS | P-ISSN: 2520-0755| E-ISSN: 2522-3003 | Vol. 9 No. 2 July – Dec 2025
54
 Nareta Kataria Khalid Rasheed, AI-Driven Recruitment through CV Scanning and AI Agents (pp. 54 -61)

development of an AI-based agent that can experiment based on realistic steps in

grade and assess the CV and states the recruitment decision making with
operational, technical, and ethical challenges limited information.
and emphasizes how this tool can enhance the ● Shows how supervised learning can be
hiring process. In addition to discussing performed using HR-inspired scoring
methods for improving equity, minimizing functions in the absence of labelled
bias, and preserving transparency, it describes recruitment data.
the process for data preparation, model ● Gives empirical data of predictive
training, and evaluation. By doing this, the
stability and operational efficiency
study advances the more general goal of
besides addressing ethical risks
developing hiring practices that are not only
effective but also fair, dependable, and long- including bias and interpretability.
lasting.
2. LITERATURE REVIEW
B. Problem Statement The potential of artificial intelligence (AI)
Although there are growing efforts to use to improve efficiency, consistency, and
AI in the hiring process, the majority of the quality of decisions in information
literature on the topic addresses single steps of management and human resource processes
the process (e.g. resume parsing or interview has sparked interest in its use in recruitment
analysis) or it is based on large proprietary and hiring. Researchers have investigated
datasets that are not available in practice in various AI-based recruitment tools, such as
any organization. As such, very little resume filtering technologies, interview
empirical data exists regarding the analysis, candidate ranking systems, and
effectiveness of the end to end AI-driven decision-support tools. A significant portion
recruitment systems in the confines of small of available literature discusses recruitment
samples and the real-world environment and processes without acknowledging the
in terms of fairness, transparency, and automated recruitment pipeline as a whole,
predictive accuracy. This study aims to fill this considering its various components in
gap by exploring an AI-based approach to isolation. Initial research centered on attempts
recruitment framework that is scenario-driven to automate resume screening using simple
and tested with one of the restricted real CV keyword matching via rules or via automation
data. learning using skill extraction [1] summarizes
analytically the recruitment automation
C. Research Objectives literature and discusses the constructive
● Explore AI-driven CV scanning for automation of recruitment as an approach to
automated candidate screening. reduce the burden of the recruiter and elevate
● Analyze AI agents for scheduling and the speed of the process. Though the case
candidate communication in recruitment remains conceptual with no design or
automation. operational recruitment system framework to
● Assess the effectiveness of AI in evaluate. Analogously, traditional supervised
reducing hiring time and bias. learning for classification and regression of
decision-support systems has been widely
D. Research Contributions covered in the literature, especially in the
● The research contributes the following: foundational pieces [2], [3] but still remains
● Suggests a complete AI-based largely unutilized to recruitment in a real case
recruitment system combining the scenario with limited data. Recent studies
parsing of CVs, machine learned have incorporated AI into later phases of
candidate ratings, and AI-agent mediated recruitment, especially interview assessments.
communication. [4] created an AI remote interview system that
analyzes speech, tone, and facial features, and
● Presents an experimental assessment of a
works in remote interviews which show strong
scenario that is done in the form of correlation with human evaluators. While
Sukkur IBA Journal of Computing and Mathematical Science - SJCMS | Vol. 9 No. 2 July – Dec 2025
55
 Nareta Kataria Khalid Rasheed, AI-Driven Recruitment through CV Scanning and AI Agents (pp. 54 -61)

such systems help improve evaluations, they applicability in small and medium-sized
do not fix the issue of high-volume resume organizations that operate in data-scarce
filtering, as they sit beyond preliminary environments and are focused on medium-
candidate screening. Pathak and Pandey [5] and small-scale operations. Thirdly, the
offer multi-agent recruitment systems that literature still lacks discussion on the ways
automate interview scheduling and candidate that proxy and/or HR-style target variables
assessments, yet their work is agent-centric, may be strategically and systematically used
assuming access to extensive or simulated to facilitate supervised learning in the absence
datasets, and largely neglects score-based of labeled outcomes in the recruitment
resume filtering. Other than technical process. This study fills these gaps with the
frameworks, some researchers have focused first end-to-end AI-driven recruitment
on the organizational, ethical, and automation framework, fully designed and
psychological aspects of AI-aided empirically tested under plausible data
recruitment. [6] formulated a decision-making limitation constraints. In contrast to previous
model on the adoption of AI interviewing research focusing on specific recruitment
systems, based on its perceived utility and components or conceptual models of
ease of use. Regarding AI recruitment, [7] adoption, this study focuses on scenario-
studied candidates’ affective responses and based/mobile evaluation, predictive stability,
noted that acceptance is contingent on the and operational viability employing a
recruitment process being transparent and fair. restricted real CVs set. By combining machine
On the other hand,[8] raised concerns and learning-based scoring of candidates with AI-
underscored the issues of accessibility and agent assisted candidate interaction and
fairness, especially in the context of the addressing the ethical concerns of bias,
disabled. While the studies mentioned are explainability, and other associated risks, this
pivotal in understanding the phenomena of study provides empirical evidence to bridge
trust, bias, and acceptance, none of them the gap between the theoretical and the
provide a predictive analysis of the AI practical components of AI recruitment
recruitment systems in question. The latest research and automation hiring.
studies have investigated the impact of AI on
sustainability and efficiency in HR processes. 3. METHODOLOGY
[9] demonstrated that generative AI in This research employs a scenario-driven
recruitment can lower the consumption of experimental approach to design, implement,
administrative resources and overhead, and and assess an AI-based framework for resume
their emphasis is on process efficiency rather screening and recruitment automation. The
than candidate assessment. [10] studied the way it is structured is as a set of interconnected
impact of personality traits and the acceptance processes instead of a set of procedural steps
of recruiting AI on the adoption of AI- to be followed, supported by three realistic
assistant hiring, which strengthens the case for recruitment situations, hence making sure that
users-centered system design. Regardless of the system evaluation is a reflection of the real
these developments, many gaps still exist life hiring conditions.
within the literature. Firstly, a small number of
studies provide detailed, comprehensive, and A. Data Acquisition and Scenario Design
holistic recruitment frameworks that integrate There were 50 real CVs as PDF files
components such as resume automated collected and used as the main data set. An
parsing, candidate scoring, decision isolated Job Description (JD) of a technical
thresholding, and interactive AI recruitment position was developed. The system was
chatbots/agents within the same system. evaluated using the following three scenarios
Secondly, existing empirical assessments are of recruitment according to the supervisor
often based on proprietary data and/or large orders:
datasets, which affects the empirical
assessments’ reproducibility as well as its
Sukkur IBA Journal of Computing and Mathematical Science - SJCMS | Vol. 9 No. 2 July – Dec 2025
56
 Nareta Kataria Khalid Rasheed, AI-Driven Recruitment through CV Scanning and AI Agents (pp. 54 -61)

● Scenario 1 (Bulk Screening): The whole CVs) is realistic pilot scale recruitment
50 CVs are done simultaneously to scenarios as opposed to benchmark optimal
simulate mass recruitment. model optimization. This study does not aim
● Scenario 2 (Shortlisting to Preliminary to make a large-scale generalization but rather
Interview): Ten most successful to perform a controlled test of viability,
candidates are selected to undergo the stability, and consistency of decisions within
preliminary screening process. the constraints of operations. Since labelled
● Scenario 3 (Final Selection): The top recruitment datasets were not available, an
artificial HR-inspired target score was created
three candidates are shortlisted to the
which mimics human measuring:
final interview process.
● Such cases directly align with the real ● Score = (Years of Experience × 10) +
HR recruitment processes and allow (Number of Skills × 5) + (Education Level
presenting the results in a practical and × 5)
decision-oriented way. This formulation represents pragmatic HR
A. Text Extraction and Feature scoring logic, and it allows guided learning
Engineering within realistic conditions.
In order to extract unstructured textual content C. Model Selection and Training
in CV PDFs a text extraction pipeline based Various regression models were under
on the PyMuPDF (fitz) library is used. The experiment such as Linear Regression,
text extracted is processed to create rule-based Support vector Regression (SVR), Gradient
and pattern-matching processing in order to Boosting Regression (GBR) and Random
generate structured numerical features. Forest Regression. These selection criteria
● Years of Experiences: Calculated with the were:
help of the regular expression analysis.
● Predictive accuracy (R²)
● Number of Technical Skills: Decided with
● Error in prediction (Mean Absolute Error -
the help of a given skill lexicon.
MAE)
● Education Level: Was represented in
● Robustness to noise
numerical terms based on qualification
● Capability of representing nonlinear
(Diploma, Bachelor, Master, PhD).
interactions of features.
● Skill Overlap with Job Description:
● Suitability for small datasets
Identified based on the comparison of CV
skill with the skills needed in the Job Since all models tested showed some
Description. degree of instability, the Random Forest
● Coverage Ratio: The coverage ratio was Regressor showed the best and most
determined by dividing skills on the job consistent performance. The model was then
description, which are found in each trained on a 8020 train test split of 300
curriculum resume. decision trees. The trained model achieved:
● These characteristics constitute predictive
input variables of machine learning based ● R² = 0.94
scoring. ● MAE = 4.46

B. Target Variable Construction These findings confirm the high

Because there are no publicly available correlation between AI predicted scores and
labeled datasets of recruitment, a scoring HR-inspired ground truth scoring.
function inspired by HR has been developed
to provide a linkage to actual screening. These
proxy labeling approaches are typically
pursued in applied machine learning studies in
data scarcity situations. The data size (50
Sukkur IBA Journal of Computing and Mathematical Science - SJCMS | Vol. 9 No. 2 July – Dec 2025
57
 Nareta Kataria Khalid Rasheed, AI-Driven Recruitment through CV Scanning and AI Agents (pp. 54 -61)

D. Scenario Based Evaluation and Fig 1: The scatter plot demonstrates that the
Categorization highest accuracy (R²) is obtained with the least
All the CVs have been rated and divided error (MAE) with the help of Random Forest
by the specified recruitment scenarios using which means that the model has the best
the trained Random Forest model: predictive stability.
● Category A: All candidates (initial
screening pool)
● Category B: Top 10 candidates (first
interview round)
● Category C: Top 3 candidates (final
interview stage)

The ranked predictions were automatically

used to compute dynamic score thresholds to
make selections adaptive to various job roles.

E. Interconnecting with AI Agent Fig. 2. The heat map algorithms metric

Interface. dimensions.
The trained model is incorporated in a
graphical HR application (Tkinter based GUI)
that permits:
● Categorizing candidates automatically.
● Prediction of uploaded CVs in real time.
● Exporting of shortlisted candidates into
CSV files to be used by the HR.

This integration will make sure that the

methodology of the experiment is not merely
a theory, but a theory that is actually proved in
a functional recruitment automation system.

Fig. 3. Stacked bar comparison of model

performance (accuracy, error, and
computational cost.

Fig 3: The stacked bar plot shows that the

best-balanced performance based on
performance in terms of accuracy, error and
cost of computation is offered by the Random
Forest. The classification of the candidates is
carried out through a thresholding model,
which is forecasted through the distribution of
scores of the training set. This will enable the
division of the candidates into Category A (all
Fig. 1. Scatter plot of R² versus MAE showing applicants), Category B (top 10 shortlisted)
Random. and Category C (top 3 finalists). The
thresholds will ensure that both existing and
Sukkur IBA Journal of Computing and Mathematical Science - SJCMS | Vol. 9 No. 2 July – Dec 2025
58
 Nareta Kataria Khalid Rasheed, AI-Driven Recruitment through CV Scanning and AI Agents (pp. 54 -61)

new submissions of resumes receive equal 5. RESULTS AND DISCUSSION

classification. All the functional elements are
The large R value suggests that there is a
then brought together into a graphical user
high level of congruence between HR inspired
interface (GUI) manufactured using the
scoring and not absolute hiring correctness,
Tkinter, allowing the HR personnel to
which is in line with the aim of the study to
visualize job descriptions, search through
provide support and not replacement to the
numerous resumes, create classifications, and
decision making. Nonetheless, the obtained
predict the type of new applicants.
results demonstrate that feature weighting and
4. SYSTEM dataset composition are sensitive and need
retraining on a regular basis and human
DESIGN/IMPLEMENTATION
supervision. These results support the
A. Overview literature in the area of algorithmic bias and
The system that will be proposed interpretability where it is noted that
comprises of three key stages: predictive performance is not enough to
justify ethical use of recruitment algorithms.
● CV Parsing: Get the relevant skills,
experience, and qualifications. A. Efficiency Improvements
● Candidate Scoring: ML models should be CV scanning saves a lot of time in
used to rank the candidates based on their screening the candidates by automation. The
suitability towards the job. system takes an average of 50 of the CVs in
● AI Interaction with the Agents: Interview, minutes as opposed to hours or days required
respond to questions and take feedback. in a manual screening system. This enables
HR teams to work on more strategic and
A. Workflow Diagram senior level decision making and planning as
opposed to mundane duties.

B. Accuracy and Candidate Matching

ML scoring and NLP based parsing makes
sure the candidates are judged in a similar
manner according to the predefined criteria,
including skills, experience, and
qualifications. Accuracy measures show that
there is a high level of precision in the
identification of appropriate candidates, and
false negativity will be minimized whereby
good candidates may be missed.
Fig. 4. Workflow of the proposed
recruitment pipeline:
C. Bias Reduction
The system reduces the effect of
B. Technical Details unconscious bias because the criteria used to
do evaluations is standardized. Rather than the
● Text extractors like PyMuPDF or spaCy. irrelevant considerations of gender, age, or
● ML classifiers (Random Forest, SVM) to ethnicity used by human evaluators, AI
rank the candidates. scoring encourages the hiring decision to be
● AI agents implemented using fairer. Fairness is also achieved through
conversational AI frameworks for continuous monitoring and retraining of
interaction (See Fig. 4). models.
Sukkur IBA Journal of Computing and Mathematical Science - SJCMS | Vol. 9 No. 2 July – Dec 2025
59
 Nareta Kataria Khalid Rasheed, AI-Driven Recruitment through CV Scanning and AI Agents (pp. 54 -61)

D. AI Agent Performance project fills this gap by providing a

AI agents enhance the engagement of the comprehensive, end-to-end automation
candidates by making automated pipeline to deal with resume processing and
communication, arranging the interviews, and shortlisting. The system uniquely combines
delivering prompt feedback. This lowers the resume text extraction, structured feature
administration burden and candidate engineering, job description-based skill
experience is enhanced. Such metrics as the alignment, multi-model evaluation, Random
response rate and satisfaction questionnaires Forest predictive scoring, threshold-driven
are positive, and they indicate that interactive categorization, and a GUI-enabled
AI is useful in recruitment. In combination, deployment for HR professionals. This project
CV scanning and AI agents have an enormous provides a practical and executable solution
recruiting efficiency that neither of the that directly improves early-stage recruitment
strategies can achieve independently. Such efficiency and decision-making, in contrast to
limitations are the use of non-standard CV the primarily conceptual or interview-based
forms, possible excessive dependence on AI studies examined. This positions the
scoring, and a need to upgrade the model developed system as a pragmatic adjunct to
regularly. Real life implementation offers current research, connecting theoretical AI
measurable benefits to both operational recruitment frameworks with practical resume
effectiveness of HR and satisfaction of automation tools.
applicants and can be highly applicable in a
ACKNOWLEDGMENT
business environment.
Most sincerely I would like to thank the
6. CONCLUSION instructor for all the help, motivation, and
encouragement given in the course of this
work. Their valuable comments, practical
The comparison of the literature shows valuable suggestions and profound knowledge
that the current studies of AI-based recruiting were useful and necessary for defining the
solutions are concentrated mostly on course and the degree of this production.
individual aspects of hiring rather than Therefore, I would like to be grateful to them
offering a unified system. Without providing for the part they have provided in the creation
a practical framework, Albassam (2023) of this research and for their passion and
provides a conceptual overview of AI-driven readiness to pass knowledge which have
recruitment strategies. Similarly, Lee and Kim become my motivation during work on this
(2021) and Pathak and Pandey (2025) focus paper and their supervision. Thanks for being
on AI-improved interviewing platforms and so understanding with us, for enriching our
multi-agent evaluation systems and install passions, and for expecting the best from us.
their contributions at a later stage in the
REFERENCES
recruitment process. Other works in the
literature, including the models of [1] W. A. Albassam, “The power of artificial
organizational decision making and adoption intelligence in recruitment: An analytical
(Lee & Kim, 2021), sustainability and process review of current AI-based recruitment
optimization through generative AI usage strategies,” International Journal of
(Koteczki et al., 2025), the role of affective Professional Business Review, vol. 8, no. 6, p.
responses to AI usage in candidates 4, 2023.
(Koechling et al., 2023), concerns of fairness [2] B. Liu, Web data mining: Exploring
and accessibility, (Fisher et al., 2024), and hyperlinks, contents, and usage data. New
technology acceptance factors (Calluso et al., York, NY, USA: Springer, 2007.
2025), have a valuable theoretical and
contextual value but do not offer The given
Sukkur IBA Journal of Computing and Mathematical Science - SJCMS | Vol. 9 No. 2 July – Dec 2025
60
 Nareta Kataria Khalid Rasheed, AI-Driven Recruitment through CV Scanning and AI Agents (pp. 54 -61)

[3] S. B. Kotsiantis, I. Zaharakis, and P.

Pintelas, “Supervised machine learning: A
review of classification techniques,” in [9] R. Koteczki, D. Csikor, and B. E. Balassa,
Emerging Artificial Intelligence Applications “The role of generative AI in improving the
in Computer Engineering, vol. 160, no. 1, pp. sustainability and efficiency of HR
3–24, 2007. recruitment processes,” Discover
Sustainability, vol. 6, no. 1, pp. 1–28, 2025.
[4] B. Lee and B. Kim, “Development of an
AI-based interview system for remote hiring,” [10] C. Calluso and M. G. Devetag, “The
International Journal of Advanced Research in impact of technology acceptance and
Engineering and Technology, vol. 12, no. 3, personality traits on the willingness to use AI-
pp. 654–663, 2021. assisted hiring practices,” International
Journal of Organizational Analysis, in press,
[5] G. Pathak and D. Pandey, “AI agents in 2024.
recruitment: A multi-agent system for
interview, evaluation, and candidate scoring,” [11] Y. Zhang and B. C. Wallace, “A
unpublished, 2025. sensitivity analysis of (and practitioners’
guide to) convolutional neural networks for
[6] B.-C. Lee and B.-Y. Kim, “A decision- sentence classification,” in Proceedings of the
making model for adopting an AI-generated Eighth International Joint Conference on
recruitment interview system,” International Natural Language Processing (Vol. 1: Long
Journal of Management, vol. 12, no. 4, pp. Papers), pp. 253–263, 2017.
548–560, 2021.
[12] A. E. Hoerl and R. W. Kennard, “Ridge
[7] A. Koechling, M. C. Wehner, and J. regression: Biased estimation for
Warkocz, “Can I show my skills? Affective nonorthogonal problems,” Technometrics,
responses to artificial intelligence in the vol. 12, no. 1, pp. 55–67, 1970.
recruitment process,” Review of Managerial
Science, vol. 17, no. 6, pp. 2109–2138, 2023.
[8] S. L. Fisher, S. Bonaccio, and C. E.
Connelly, “AI-based tools in selection:
Considering the impact on applicants with
disabilities,” Organizational Dynamics, vol.
53, no. 1, p. 101036, 2024.

Sukkur IBA Journal of Computing and Mathematical Science - SJCMS | Vol. 9 No. 2 July – Dec 2025
61