Built an edge AI robotics platform that captures real-world environments with Gaussian Splatting and homography to generate digital twins. Trained a reinforcement learning navigation model over 50+ iterations, used Qualcomm AI Hub for Snapdragon NPU on-device inference, and deployed Sim2Real Unity simulations to Arduino Uno Q hardware for offline autonomous navigation.

Hackathon-winning web application transforming image collections into narrated digital diaries. Implemented client-side canvas image compression achieving 80% smaller payload sizes, and optimized prompt engineering reducing token usage by 25%.

Built a voice-driven AI agent helping rural citizens access financial literacy, government schemes, and grievance services through natural conversation. Integrated LiveKit, Sarvam AI, automated dialect detection, contextual RAG, semantic search, verified government APIs, and a low-bandwidth architecture designed for telephony networks.

Shipped a cross-platform nutrition tracker serving 50+ active users with natural language and camera-based meal logging. Integrated a multimodal LLM pipeline with automatic fallback, reducing manual logging time by 85%, and built a local-first SQLite data layer with encrypted JSON backups for user privacy.

Built the Bharat Governance Simulator to provide real-time impact analysis for the One Nation One Election (ONOE) initiative. It leverages HLC and Law Commission parameters to generate simulation dashboards, analyze mid-term risks, and run Monte Carlo simulations for governance impact.

A calendar website that converts Excel sheets into a clean React-based timetable. I fixed bugs and implemented a feature to import events to users' calendars. This feature was appreciated by seniors and proved genuinely useful, which gave me the confidence and motivation to keep contributing to real-world projects.

End-to-end deep learning project classifying red blood cell microscopy images as Parasitized or Uninfected. Uses a fine-tuned MobileNetV2 CNN achieving ~95% accuracy. Features a production-ready FastAPI backend with Grad-CAM visual explanations.

Developed a full-stack web application simulating real-time stock trading using Python Flask framework. The application enables users to register accounts, look up current stock prices via API integration, buy and sell stocks with virtual currency, and maintain transaction history.