Skip to main content

Project: Enterprise RAG Copilot

  • Difficulty: Intermediate to advanced
  • Primary stack: TypeScript API, Python ingestion pipeline
  • Estimated duration: 2 to 3 weeks
  • Primary hiring signal: retrieval system design
  • Primary monetization signal: internal knowledge assistant

Problem statement​

Employees waste time hunting through fragmented documentation. The copilot should answer from trusted sources with citations.

Project implementation​

Project implementationP02 — Enterprise RAG Copilot
Runnable

Stack

TypeScriptExpressHybrid retrievalZod

Folder

Open project folder

What it proves

You can build a local-first RAG pipeline with ingestion, retrieval, citations, and eval-driven validation before introducing vector database complexity.

Next upgrade path

Add embeddings, a vector database, and generation through Project 01.

Key commands

npm ci
npm run typecheck
npm run build
npm run smoke
npm run eval

Core workflows​

  • ingest documents with metadata
  • create chunks and embeddings
  • run hybrid retrieval and reranking
  • generate grounded answers with citations
  • evaluate retrieval and answer quality

Architecture​

Milestones​

  1. Ingest a small document set with metadata
  2. Add hybrid retrieval and citation formatting
  3. Add reranking and refusal behavior for missing evidence
  4. Add eval set and dashboard metrics

Acceptance criteria​

  • every answer cites one or more source chunks
  • the system refuses unsupported claims
  • retrieval can be debugged with source metadata
  • at least one offline eval dataset exists

Starter implementation​

Starter code is now available in projects/p02-enterprise-rag-copilot. The current starter uses local JSON storage plus deterministic hybrid scoring so learners can understand the pipeline before adding embeddings or a vector database.

The starter also includes a minimal eval dataset in evals/questions.json so learners can measure retrieval quality before adding more advanced ranking and generation layers.

Portfolio packaging​

Publish screenshots of answers with citations, retrieval debug views, and a diagram of the ingestion pipeline.

Monetization path​

This is directly monetizable as a departmental copilot, enterprise pilot, or ingestion-and-search accelerator.