← Back to What we can do
~/sprint/route-03 $ ask --over ./sop-corpus▍

Doc-Q&A RAG

Demo · synthetic data

Accurate, cited answers over your own documents — and a flat refusal when the answer isn't in them. No hallucination.

retrieve · cite · answer
5 docs · 14 chunks indexed
The problem

Your answers are buried in PDFs, wikis, and tickets. People re-ask the same questions, get stale or guessed answers, and trust erodes — or an LLM invents a confident wrong answer with no source to check.

What the automation does

Retrieves the most relevant passages from your own docs and answers only from them, every claim cited. Out-of-scope questions get a refusal, not a guess. Here: top match 0.96, zero unsupported claims.

Suggested questions
>
Keyword retrieval over an embedded corpus — no API, no keys. Try the highlighted chip to see the no-hallucination guard fire.
Retrieved context top-k = 4
No query yet. Pick a suggested question or type your own — the top matching chunks (with similarity scores) appear here, and the answer cites them by number.
Synthesized answer grounded · inline citations
The answer is generated only from the retrieved chunks on the left. Every claim links back to a numbered source — click a [1] to jump to its chunk.
embed query retrieve top-k rank by score answer + cite

The corpus (synthetic — 5 docs)

Invented operations SOPs & policy docs for a fictional 3PL/e-com brand. No real company, people, or values.