Retrieval-Augmented Generation System for Research Resources
End-to-end RAG application with hybrid retrieval, reranking, and a streaming FastAPI backend.
An end-to-end RAG application covering the full pipeline: web crawling and document ingestion, heading-aware chunking, hybrid dense + BM25 retrieval with reciprocal rank fusion, and cross-encoder reranking. Served via a FastAPI backend with streaming responses, supporting local LLM inference with cloud API fallback.
Tech stack: Python, FastAPI, hybrid retrieval (dense + BM25, reciprocal rank fusion), cross-encoder reranking, local LLM inference, streaming APIs
GitHub: [GITHUB LINK TBD]