02 / Projects
NSure-AI
Insurance Document RAG
QuantSub-50msHackathon Win95% Speedup
Problem
Insurance PDFs mix tables, clauses, and scanned pages. Naive chat-over-docs is too slow and too imprecise for underwriting questions.
Approach
Hybrid BM25 + vector retrieval on PyMuPDF-extracted text, served through FastAPI with Gemini generation and a Postgres-backed cache.
Impact
Sub-50ms retrieval, a 95% cache speedup, and a hackathon win for financial document analysis.
Stack
PythonFastAPIRAG SystemsGemini APIPostgresPyMuPDF