Things I've built. Every project here is my own work — no forks, no templates.
Built a RAG system indexing 11K+ chunks with hybrid retrieval (BM25 + semantic vectors), achieving 79% recall at sub-100ms latency. Implemented Reciprocal Rank Fusion and automated Precision@K/Recall@K evaluation; generated grounded answers with inline citations via Groq API.
Local RAG assistant with Qdrant vector database, BGE-M3 embeddings, and Ollama for LLM inference. Upload PDFs, DOCX, or TXT files and get cited answers from your own knowledge base.
Trained a CNN on 5,000 chest X-ray images to detect pneumonia (AUC 0.91) and validated results using clean evaluation splits. Containerized the model behind a Flask REST API for real-time inference.
Responsive single-page app consuming the OpenWeather API with caching to reduce repeat calls. Added input validation and a mobile-first UI for reliable searches and clean UX on small screens.
These are just the highlights.
Find more on GitHub ↗