AI System

RAG Lab

A deployed RAG backend and product surface for testing retrieval, ingestion, source behavior, and access boundaries.

Problem

Many RAG demos stop at a chat interface. RAG Lab treats retrieval as a system that should be tested: questions, chunks, sources, scores, ingestion, and permissions all become part of the product surface.

Architecture

Go API on Render, PostgreSQL with pgvector, provider-based embeddings and generation, scoped temporary tokens, and a Next.js frontend deployed on Vercel.

Decisions

  • Expose retrieval context in the public demo because the goal is to evaluate the system, not hide it behind a generic assistant.
  • Separate safe demo capabilities from admin-only operations with scoped temporary tokens.
  • Keep the frontend closer to product documentation than chat UI, so the API, workflow, and boundaries remain visible.