RAG Nutrition App Backend Build

I’m finalising a nutrition-focused mobile product and now need the server side that can truly “think” before it answers. The stack is Node.js with Express and MongoDB, and the core requirement is an LLM pipeline that relies on retrieval-augmented generation, never hallucination.

Here is what I need you to engineer:

• A RAG workflow that pulls only verified nutritional data from my Mongo collections and feeds it to the language model, with additive analysis treated as first-class information. Accurate ingredient analysis is the single most important metric for success, so the retrieval layer has to be rock solid and testable.

• User context handling: every response must factor in stored health profiles— allergies, chronic conditions, dietary goals— so that the output includes personalised health scores and risk insights.

• Clean REST endpoints for ingredient look-up, additive breakdown, profile management, and the generated insight feed, all secured and documented.

• Clear evidence of similar AI + external-data builds you have already shipped; I’m especially interested in projects where you solved RAG grounding problems rather than calling a vanilla ChatGPT endpoint.

Deliverables
1. Source-controlled Express server (TypeScript preferred but JS is acceptable)
2. Schema-driven MongoDB collections for ingredients, additives, verified nutritional data and user profiles
3. RAG module (vector store set-up, embedding pipeline, prompt templating, evaluation scripts)
4. Postman or Swagger docs plus a short read-me on deploying and fine-tuning the system

Acceptance Criteria
• Any answer the model returns must reference a document ID from the database.
• Response latency <2 s on common queries.
• Test suite shows 0% hallucination on the provided evaluation set.

If you have shipped something similar, let’s talk—I’m ready to move quickly.