RAG AI Data Fetch Structure

My e-commerce site already holds every product record in a MongoDB cluster, and I now want a Retrieval-Augmented Generation (RAG) layer that can pull that data on-demand for search, chat, and merchandising features.

Here’s what I need built:

• A lightweight service (Python preferred; open to Node.js) that connects securely to the existing MongoDB collection and exposes a clean retrieval interface.
• An embedding/indexing routine so product documents can be vectorised and queried efficiently (e.g., with FAISS, Pinecone, or a native MongoDB Atlas Vector Search).
• RAG pipeline logic that combines the retrieved product fields with a language model prompt to produce rich, real-time answers.
• Environment configuration and clear instructions so I can run, extend, and monitor the service in staging and production.

Acceptance criteria
• Given a product ID or free-text query, the service returns the correct name, price, and description in under 500 ms.
• All queries must hit only the internal MongoDB instance—no external data sources.
• Codebase passes a quick hand-over review (readme, sample .env, and one command to start).

Let me know which stack you prefer, any prior RAG implementations you have, and an estimated timeline to get a first working prototype in my hands.