Projects
Gaussian Splatting SLAM with ORB-SLAM3 Tracking — CMU, Course Project
Combined ORB-SLAM3 tracking with Gaussian Splatting for accurate localization and photorealistic maps (25.1 dB PSNR), cutting trajectory error 10× over standalone Gaussian Splatting SLAM on a self-collected 48 m corridor dataset.
Deep Reinforcement Learning for UR5 Pick-and-Place — CMU, Course Project
Engineered a UR5 grasping task in PyBullet and benchmarked model-free (SAC, PPO) vs. model-based RL (learned dynamics and MPC), scaling to direct joint control.
Real-Time Image and Video Dehazing Algorithm
Capstone project developing real-time dehazing algorithms for images and video. Achieved 30ms per image and 10ms per video frame using adaptive contrast-based methods and temporal coherence optimization. Published in IEEE I4C 2024.
Privacy Preserving Machine Learning using zk-SNARKS — RIT, Course Project
Built a privacy-preserving ML framework combining homomorphic encryption and zk-SNARKs to validate computations without exposing sensitive data. Achieved 95% accuracy on MNIST using PyTorch and the Zama protocol for fully homomorphic encrypted datasets.
Assistive Device for the Visually Impaired — IEEE CIS, Bengaluru, India
Engineered a low-cost wearable device with object detection, scene recognition, and multilingual support for daily navigation and reading. Integrated OCR with Text-to-Speech functionality and deployed to 100 students at the National Association for the Blind.