Introduction
Hello, I'm Manan Chvada, a third-year [Link] Mechanical Engineering student and minoring in Computer
Science and Engineering at IIT Gandhinagar. My core skills lie in programming and development, with a
strong background in problem-solving and how to utilise Artificial Intelligence to build applications for
real-world challenges. Solving over 600+ problems has given me a strong foundation in problem-solving
using various algorithms and Data structures.
Over the past years, I’ve worked on a range of technical projects. In my recent internship as an AI/ML
Research Intern at the Human AI Interaction Lab, IITGN, I built a CNN-based gaze-controlled
software that predicts the real-time screen coordinates from eye images on React Native with a Flask
backend.. The project involved collecting and processing large-scale eye-gaze datasets, training
deep learning models for accurate gaze estimation, and deploying the pipeline into a mobile
application for real-time usage. It was meant to be a solution for motor-impaired users to get access to
technology and thus bridge the gap.
In my last semester, I led a team of 5 to build a Library management system for IIT Gandhinagar as a
course project under a professor. Yogesh Meena. This helped me learn core concepts of database
handling and api integrations.
The Machine Learning course in my second year under prof Nipun Batra has given me a strong
foundation in using various machine learning techniques and has given me some hands-on experience
working with CNN for image processing, and next-token prediction using Natural language
processing concepts.
Through my internships and projects, I have gained great hands-on experience with Backend, databases,
APIs, and machine learning algorithms to build real-world projects.
During my recent internship as an AI/ML Research Intern at the Human-AI Interaction Lab, IIT
Gandhinagar, I developed a CNN-based gaze-controlled software system aimed at improving
accessibility for motor-impaired users. The system was built on React Native (for mobile front-end) with a
Flask backend (for model inference).
The workflow involved capturing eye images through the mobile camera, which were then transmitted
to the backend where a Convolutional Neural Network (CNN) was trained to predict the user’s gaze
coordinates on the screen. These predicted coordinates were used to control on-screen elements,
enabling users to interact with applications and navigate interfaces using only their eye movements.
Resume
- My most significant internship has been at Haix Lab iitgn, where i developed a cross-platform
gaze-controlled system using react native and Flask backend with a deep-learning model. We
have also developed an authenticated API Key for all the developers who can use our model to get
coordinates and develop something meaningful.
- Testing on various models from CNN to Alexnet to get better on-device predictions. It was
meant to be a solution for motor-impaired users to get access to technology and thus bridge the
gap. We have also developed an authenticated API Key, and are planning to open-source the
model for all the developers who can use our model to get coordinates and develop something
meaningful.
-
- In the Library management system, another project I led a team of 5 to build a Library
management system
- Entity relationship diagram for workflows.
- Optimised SQL queries for fetching results.
- 10-12 tables linked using foreign keys and primary keys.
- Added Email notifications
- CNN for image super resolution and reconstruction, to understand next-token prediction over
Natural language processing concepts.
-
Why should we hire you?
- I think basically the skills I’ve gained by having a strong foundation in development on working on
various projects, and building scalable and secure systems in journey till day, aligns perfectly with
this role. My Passion towards developing solutions to real-world problems motivates me to work
consistently, learn consistently and grow consistently.
Motivation?
My primary motivation is the challenge of solving complex, real-world problems. The opportunity to work
on a platform that impacts millions of businesses and contributes to the financial infrastructure of India is
incredibly exciting to me. I am motivated by providing the features that will be used directly by the
people, something that really matters to me, thereby giving me an identity of representing the
company with my skills and the contributions I will make. I am driven to learn and grow as a software
engineer, for which I believe Razorpay offers the best platform and environment for me to contribute, learn
and more importantly, grow.
Strengths
- Along with my expertise in building applications, I believe my strengths include problem-solving
ability. Throughout my journey, solving such problems has trained me to break down complex
challenges into simple executable steps,
- I believe in pre-planning before execution, to keep the deadlines in my mind and avoid last-minute
chaos. I actually don't love chaotic life, so I try to keep my work as well as my life as structured as
possible.
- I learn from my failures and mistakes.
- Secondly, I think my management skills help me to perform better than other, especially when we
have to handle academic life with personal growth. Time management and punctuality have always
made things easier for me since my JEE preparation time, and I am consistently following through.
Weaknesses
So, one of my weakness, On which I’m consistently working to improve my leadership skills,
I always fear of
I try to do everything by my own, I prefer to not to rely on others making my less socialise in return, which
I know very well is not a good sign, but I am constantly improving this issue with building trust among the
team, and understanding others strengths and having confidence in people gradually.
"Describe a time you handled conflict”
Where do you see yourself in 5 years?
Genuinely speaking, in 5 years I see myself as a proficient, successful software engineer, ideally at a
company like Meetmux, leading some innovative, large-scale projects.
Questions
- "What opportunities are there for an intern to learn new technologies and grow their skills beyond
the immediate project?"
- "Can you tell me more about the mentorship or team structure for interns?"
-
🔹 1. Problem / Motivation
“Many motor-impaired users face difficulties accessing technology because conventional input devices like
keyboards and touchscreens are not usable. I wanted to build a system that allows them to control apps
and interact with a device using only their eye movements.”
🔹 2. Objective
“The goal was to create a gaze-controlled software that could take real-time camera input, detect the
user’s gaze, predict where they are looking on the screen, and use that to control on-screen elements.”
🔹 3. Data Collection
“To achieve this, I first needed data. I built a React Native mobile app that showed a red dot moving
across different screen points (like a 3×3 or 9×9 grid). At each point, the app captured images of the
user’s eyes and stored them with the corresponding screen coordinates. This formed the dataset for
training.”
🔹 4. Model Development
“Using this dataset, I trained a Convolutional Neural Network (CNN) with TensorFlow/Keras to map eye
images to screen coordinates (x, y). I experimented with architectures like MobileNet, ShuffleNet, etc., and
compared results based on metrics like Mean Absolute Error (MAE) and R² score. The best-performing
model was then chosen for deployment.”
🔹 5. Backend & Integration
“The trained model was deployed on a Flask backend server, which received eye images in real time
from the mobile app. The backend processed the images, predicted gaze coordinates, and sent them
back to the mobile app.”
🔹 6. Frontend (User Interaction)
“On the mobile app side (React Native), the predicted coordinates were used to highlight or select
elements on the screen. For example, in a prototype gaze-controlled keyboard, users could type letters
just by looking at different regions of the screen.”
1. Problem / Motivation
“In my project, I wanted to build a gaze-controlled software that could help motor-impaired users
interact with technology without needing touch or physical input devices.”
2. Challenge (No Dataset Available)
“One of the biggest challenges was that no suitable public dataset was available for mobile
eye-gaze prediction. So, I decided to create my own dataset collection pipeline.”
3. Dataset Creation
● “I built a React Native mobile app that displayed a moving red dot across different points on
the screen (like a 3×3 or 9×9 grid).”
● “At each position, the app captured images of the user’s eyes using the front camera at 30
frames per second (30 FPS).”
● “Each captured image was automatically tagged with the corresponding screen coordinates
(x, y) of the red dot.”
● “This gave me a labeled dataset of eye images paired with gaze positions.”
4. Model Development
● “I trained a Convolutional Neural Network (CNN) using TensorFlow/Keras to predict the (x,
y) screen coordinates from eye images.”
● “I experimented with multiple CNN architectures (MobileNet, ShuffleNet, etc.) and evaluated
them using metrics like Mean Absolute Error (MAE) and R² score.”
● “After fine-tuning, I picked the best-performing model for real-time use.”
5. Backend & Integration
● “I deployed the trained model on a Flask backend.”
● “The React Native app continuously sent captured eye images to the backend, which
predicted gaze coordinates and sent them back.”
6. Real-Time Application
● “On the frontend, I integrated these predictions into interactive applications. For example, I
built a gaze-controlled keyboard where users could select letters by simply looking at
specific regions of the screen.”
● “This made the system usable in real time and responsive enough for practical interaction.”
7. Results & Impact
● “The system successfully predicted gaze coordinates in real time with reasonable
accuracy.”
● “This demonstrated that even without an existing dataset, we could build an end-to-end
gaze-controlled interface for assistive technology.”
● “The project has strong potential to make devices more accessible for motor-impaired
users.”
