GiftNutriPepper — Food Safety & Nutrition AI
General-purpose LLMs give generic, sometimes inaccurate food safety advice that doesn't reflect local guidelines or verified nutritional standards. This is dangerous in a food safety context.
Built a constrained AI system using FastAPI, ChromaDB, and MPNet embeddings that only answers from a curated knowledge base of verified food safety and nutrition documents. The RAG pipeline retrieves the most relevant chunks before any generation — keeping the LLM honest and on-topic.
Every response is grounded in verifiable, localised sources. Users get trustworthy food safety guidance without the hallucination risk of a general-purpose AI.
Constraining an LLM is as important as prompting it. The retrieval layer isn't just a performance optimisation — it's a safety mechanism. The architecture decision to gate generation through retrieval was the most important design call in the whole project.