Multi-Target, Multi-Camera Tracking — Summer Research Internship, IIT Bombay (May 2025 – July 2025)

Advisor: Prof. Ganesh Ramakrishnan

Description: Developed a 3D object detection and multi-target tracking system for multi-camera setups with depth sensors on AICITY 2025 synthetic datasets, addressing challenges like occlusion and variable lighting.

Results: Enhanced cross-camera identity matching using geometric constraints, 3D pose estimation, and spatio-temporal modeling. Work is progressing toward a research publication.

Technologies Used: Python, PyTorch, OpenCV, NumPy, 3D Vision, Multi-view Geometry

Animal Pose Transfer and Regeneration (August 2025 – Present)

Advisor: Prof. Shanmuganathan Raman

Description: Developing a diffusion-based pipeline to generate 3D quadruped avatars from monocular images and enable pose transfer between different animal species while preserving texture and background realism.

Technologies Used: Python, PyTorch, Diffusion Models, 3D Reconstruction, Computer Vision

TinyWL Window Manager for Linux (December 2024 - Present)

Advisor: Prof. Balgopal Komarath

Description: Developing a custom window manager, TinyWL, in C, focused on adding advanced features such as dual-window display, stacking, tiling, and window merging (snapping). This project incorporates the implementation of efficient window management techniques, enhances user interaction, and contributes to the Linux Society.

Technologies Used: C, Linux

GitHub Repository: Project Link

Exploration Beyond Vanilla HNSW: Learning Graph Structure for Smarter ANN Search (February 2025 – April 2025)

Advisor: Prof. Anirban Dasgupta

Description: Improved the HNSW algorithm through frequency-based layer assignment, clustering-driven connections, and SCC-aware pruning to enhance Approximate Nearest Neighbor (ANN) search performance.

Results: Delivered higher recall and faster query times than standard HNSW on large-scale datasets.

Technologies Used: Python, NumPy, scikit-learn, Faiss, NetworkX

GitHub Repository: Project Link

Multi-Modal Fake Review Detection Engine (HackOn with Amazon) (April 2025)

Team: AWSome Hack, IIT Gandhinagar

Description: Built a multi-modal fake review detection system integrating product data, user reviews, and metadata. Used CLIP-based cross-modal validation and various data encoders for training to identify deceptive patterns.

Technologies Used: Python, PyTorch, CLIP, NLP, Sentiment Analysis, MLPs

GitHub Repository: Project Link

Re-Identification Model using Body Shape and Depth Data (July 2025 – Present)

Advisor: Prof. Ganesh Ramakrishnan

Description: Building a re-identification model based on CLIP-ReID that leverages human appearance and depth information to capture 3D body geometry for improved identification accuracy.

Technologies Used: Python, PyTorch, CLIP, Depth Estimation, Computer Vision

Inter IIT Tech Meet 13.0 — Zelta Automations Algorithmic Trading (December 2024)

Team: IIT Gandhinagar Contingent

Description: Designed and developed algorithmic trading strategies for BTC and ETH pairs as part of the Inter IIT Tech Meet 13.0 challenge. Implemented dynamic risk management frameworks to optimize trade execution. Performed extensive multi-year backtesting and validation to ensure robustness across volatile market conditions, achieving a Sharpe ratio of 2.83 on ETH and 1.96 on BTC.

Technologies Used: Python, Pandas, NumPy, Backtesting.py, Technical Analysis Indicators, Machine Learning

GitHub Repository: Project Link

ES204 Final Project — Digital Systems (Open-Source Project) (March 2025 – April 2025)

Advisor: Prof. Joycee Mekie

Description: Designed a 3×3 systolic-array accelerator in Verilog with pipelined multiply-accumulate operations for efficient matrix multiplication using streaming dataflow.

Implementation: Integrated UART RX/TX for FPGA–host communication on Basys3 and developed a Python utility for transmitting 8-bit matrices and receiving computed results over serial.

Technologies Used: Verilog, Python, FPGA (Basys3), UART Communication

GitHub Repository: Project Link

MLP-based Test Generator (Aug 2024 - Nov 2024)

Advisor: Prof. Nipun Batra

Description: Developed and implemented a sophisticated text generation model from the ground up using a Multi-Layer Perceptron (MLP) architecture, leveraging the rich linguistic patterns from Shakespearean text for training. The model was deployed on a dynamic Streamlit app, allowing users to seamlessly customize parameters such as vector length and paragraph length, enabling personalized, high-quality text generation.

Technologies Used: Python, Streamlit, Machine Learning (MLP), Natural Language Processing (NLP)

Live Demo: Link to app

Quantitative Trading (October 2024 - Present)

Description: Developed cryptocurrency trading strategies focusing on Ethereum and Bitcoin, incorporating technical indicators, machine learning models, and risk management techniques. The model integrates technical indicators, Fibonacci retracements, dynamically changing neural networks based on the Market Regime model using GARCH, clustering, and risk management tools like covariance analysis and Value at Risk (VaR). Currently working on enhancing the model with LSTM networks and advanced quantitative finance theories.

Technologies Used: Python, Machine Learning, GARCH, LSTM, Quantitative Finance, Risk Management

GitHub Repository: Link to repo

Flask-based Data Visualization Tool (Jan'24 - Apr'24)

Advisor: Prof. Mayank Singh

Description: Developed a comprehensive Electoral Bonds Data Analysis platform, converting PDF data into dynamic visualizations. Implemented advanced search, donation insights, and interactive charts (Pie, Bar, Line) using ChartJS. Enhanced data exploration with custom visualizations and exportable plots, providing actionable insights into political donations.

Technologies Used: Python, Flask, ChartJS, Data Analysis

GitHub Repository: Link to repo

Laser-Based Wireless Communication (Date: Jan'24 - Apr'24)

Advisor: Prof. Arup Chakraborty

Description: Developed a simulation model for optical wireless communication using laser technology in Wi-Fi networks, focusing on varying frequency and phase differences to transmit data. The project explored data encoding, transmission, and reception processes using high-low signal pairs to encode alphanumeric data. Addressed challenges such as signal transition delays, external light interference, and synchronization issues.

Technologies Used: Laser Technology, Signal Processing, Data Encoding, Optical Wireless Communication

Code and Documentation: For detailed analysis, technical documentation, and the code, please refer to the linked paper: Laser-Based Wireless Communication Paper

GitHub Repository: Link to repo

Human Activity Recognition (HAR) (Date: Aug 2024 - Nov 2024)

Advisor: Prof. Nipun Batra

Description: Developed a model for recognizing human activities using accelerometer data from the UCI-HAR dataset. The project included tasks such as data preprocessing, feature extraction, and visualization of activity classes using Principal Component Analysis (PCA). The model also involved training decision trees on raw accelerometer data and extracted features to classify static and dynamic activities. Conducted performance evaluations using accuracy, precision, and recall metrics.

Technologies Used: Python, Scikit-learn, PCA, TSFEL, Data Preprocessing, Feature Engineering

Code and Documentation: For detailed analysis, technical documentation, and the code, please refer to the linked repository: HAR Project Repository