DEMO / RECORDING
Overview & result
A browser-based systematic-review assistant uses Jev to locate lines in trial reports and supplements. Code copies verbatim quotes with file/page citations, and the researcher checks them before exporting an evidence table.
From the original post
Every answer is a **verbatim quote** with its file and page,
How Jev fits in the loop
- Parse local study files into line, paragraph and table-row candidates with locations.
- Use Jev Choice to screen candidate lines and Noul questions to verify whether a line answers the extraction prompt.
- Copy exact source text and location into the browser UI; require researcher review before accepting it.
How to reproduce
- Open the public demo or repository sample study in a browser.
- Ask a trial-report extraction question and inspect the highlighted quote, file and page location.
- Check the quote manually before adding it to an exported extraction table.
Why this build matters
Shows Jev as a retrieval judgment layer, not a text generator: it points at evidence while source fidelity and final acceptance stay with code and a human.
Limitations
- Jev selects candidate lines; a human must verify each quote and interpret the evidence.
- Not-found or unclear is allowed when no line passes; this is not a guarantee that the source lacks the answer.