Live Demo · Document Knowledge Base (RAG)

Ask a question. Watch the system find the answer in company documents.

This is a working retrieval pipeline running entirely in your browser — every score you see below is computed live, not pre-recorded. Same engine, four different Melbourne businesses — switch industries below.

Pipeline

Understand the question

Break the query into search terms; drop filler words.

Score every document chunk

BM25 relevance ranking across the whole library.

Compose the answer from the sources

Only retrieved text is used — every claim carries a citation.

Answer

System log

Indexed documents

What's real in this demo — and what changes in production

Real: the retrieval. Tokenisation, BM25 scoring, ranking and citation mapping all run live in your browser against whichever document set you pick above. Type anything — including a question the documents can't answer — and watch how it behaves. Notice the engine underneath is identical across all four industries; only the documents change.

Simplified: the final answer here is assembled from the retrieved sentences directly. In a client deployment, Claude writes the answer from those same sources (so it reads naturally in any language), retrieval uses semantic embeddings on top of keyword scoring, and the library covers thousands of documents instead of a dozen.

Want this running on your own documents? Book a free AI audit →