📚

DocsRAG Lab

Production-style RAG · baseline retrieval on LangChain docs

— chunks

— hit@5 baseline

Connecting…

Logs

--:--:-- Ready. Submit a question to start the pipeline.

Interactive RAG pipeline

Try “How do I create a LangChain chain?” or “What is a retriever?”

Offline evaluation

Retrieval quality

Measured against a labeled LangSmith documentation question set.

Loading snapshot…

Hit@5 — Correct source in top 5

MRR@5 — How early the first relevant source appears

nDCG@5 — Ranking quality with top-rank reward

Source diversity — Unique sources in retrieved results

Recall curve

Hit rate by cutoff

Higher is better

Dataset slices

Performance by difficulty

Labeled questions

Evaluation cases

Question-level results

—

A labeled source URL appears in the first 5 retrieved chunks. MRR and nDCG reward relevant sources that rank closer to #1. This snapshot evaluates retrieval and the configured reranker on 100 source-grounded questions (70 dev, 30 held-out test). Answer generation is excluded.

Demo project

DocsRAG Lab · production-style RAG pipeline

Interactive demonstration of retrieval-augmented generation on LangChain documentation

1 / 6