BROWSER-LOCAL DOCUMENT Q&A

Find the answer.
Read the source.

Search across documents, inspect cited passages, and export an incident brief.

Runs on your deviceSemantic + keyword search30-query benchmarkNo API subscription

QUESTION

What do you need to know?

First model load can take a few minutes. Evidence-only mode skips the LLM.

Load documents or the demo case, then ask a question.

SOURCES

Retrieved evidence

Passages appear here after you ask a question.

DEMO WORKFLOW

Build the HBR-1042 brief

Five focused checks: cause, impact, decision owner, current rollback rule, and prevention. Each finding keeps its own sources. Designed for the fictional HBR-1042 demo; incomplete answers remain possible.

RETRIEVAL EVALUATION

Compare retrieval methods

Reproduce a 30-query comparison on nine fictional operational documents. Includes paraphrases, an obsolete policy, and six questions with no documented answer. This measures retrieval—not LLM answer accuracy.

Run the comparison to see results from this browser.

ANSWER EVALUATION

Check the returned answers

Run ten actual model checks: eight factual questions and two unsupported questions. See required-fact checks, citation validation, fallback routes, and every failure. This goes beyond retrieval scores; it is still a small synthetic development audit.

No answer checks run in this session.