All projects
IntermediateIntermediate~12h
RAG Assistant
Production retrieval-augmented generation over your documents
Business problem
Answer questions over a private document set with citations, confidence scores, and graceful 'I don't know' behaviour.
Architecture
UI → API → query rewrite → hybrid retrieve → rerank → LLM with cited context → response with citations.
Milestones
Track your progress through the project.
0 of 6 milestones complete
Acceptance criteria
- Answers cite correct sources
- Refuses when context is insufficient
- Eval faithfulness > 0.85
- Latency and cost budgets met
Deployment checklist
- Vector index rebuilt on data changes
- Embedding model version pinned
- Online eval sampling
- Cost dashboard
- Trace every query
- Fallback when no context found
Evaluation criteria
- Faithfulness to retrieved context
- Citation accuracy
- Refusal calibration (false positive / negative)
- Retrieval recall@5