A mix of applied ML, distributed systems, and full-stack work — each one shipped end-to-end.
CNN + ViT detection with results verified on Ethereum and stored on IPFS.
Document ingestion → chunking → embedding → retrieval → LLM, served via FastAPI.
Custom socket layer streaming video chunks between peers, BitTorrent-style.
Converts plain-English questions into executable MongoDB queries.
A classical NLP sentiment classifier trained on 1.6 million tweets.
A library web portal with a recommendation engine driven by user behavior.
Predicts which banking customers are about to leave, with 85% accuracy.