Longevity — AI Nutrition Assistant
The idea started from a simple observation: everything a nutritionist knows, they learned from documents — scientific papers, guidelines, studies. That's exactly the kind of knowledge a RAG (retrieval-augmented generation) system can consult and answer from, grounded in real sources instead of guessing.
Nutrition documents are uploaded to the backend, vectorized, and indexed in Pinecone; when a question comes in, the system retrieves the most relevant passages and asks GPT-4 to answer using only that context. The chat also runs a short intake — age, weight, height, activity level, goals — and uses that profile to personalize the guidance it gives. The one part of an in-person visit that can't be replicated is the physical exam; letting users optionally upload a file with fuller biometric data for an even more accurate result is the next planned refinement.
The FastAPI backend and the Next.js chat frontend are deployed separately on Render and Vercel.
Designed and built the full stack solo: the retrieval pipeline, the grounding prompt strategy, and the chat frontend.