Projects

AI Bioinformatics Platform

S-BIOS

S-BIOS is an AI-based bioinformatics platform for modern analysis of imaging-based data. It is designed around workflows such as cell typing, gene expression prediction, and AI-assisted exploration of biological images.

• Supports imaging-centered analysis workflows
• Targets cell typing and gene expression prediction use cases
• Built around practical AI application needs in bioinformatics

s-bios.com

Human-Centered AI Application

BloomScroll

BloomScroll is an app aimed at addressing doom scrolling by giving users more agency over their feed. The goal is to help people receive the content they actually want or need, while making the experience playful and motivating through gamified interaction.

• Personal feed control with intentional content selection
• Gamified experience designed to make healthier scrolling feel rewarding
• Built around current problems in algorithmic content consumption

Applied Deep Learning

Deep Learning AI Proctoring Framework

A remote proctoring framework that used deep learning and computer vision methods for monitoring exam sessions, including headcount detection, gadget detection, gaze tracking, head-pose estimation, and suspicious task monitoring.

• YOLOv5 for extra person and gadget detection
• Facial landmarks for gaze, head-pose, and speaking cues
• Hackathon-winning solution at Hack-With-MAIT and Bit-by-Bit IIIT-B

Media Forensics

Deepfake Detection

A deepfake detection project using EfficientNet and engineered features such as blinking rate, frame correlation, and video quality indicators.

• Used visual and temporal features for detection
• Achieved 90%+ accuracy on standard datasets
• Implemented with Python, NumPy, OpenCV, and TensorFlow

Reinforcement Learning

Chrome Dino AI

A reinforcement learning project built around a Pygame version of the Chrome Dino game. Models were trained using game state values such as obstacle distance and achieved strong play within a small number of generations.

• Pygame implementation of the game environment
• Model training from in-game state values
• Reached four-digit scores in fewer than 50 generations

Applied Machine Learning

Fairgrades

Fairgrades explored an ML-based approach to projecting and awarding grades in exam cancellation scenarios. The project collected academic result data, converted PDFs into structured tables, and trained models to estimate student outcomes.

• Processed result PDFs into analysis-ready dataframes
• Used machine learning models for grade prediction
• Best model reported R2 score of 0.88