Real demo recording · English narration and captions · Fictional data
THE PROBLEM
When procedures and manuals are scattered, finding an answer is only the beginning: your team needs to check its evidence.
THE WORKFLOW
Query fictional documents, review document and page references, inspect the original passage and export an answer. Without enough evidence, the system abstains.
01Private documents
02Search and retrieval
03Referenced answers
04Evidence review
RECORDED EVIDENCE
Results, with context.
Results on synthetic data
Check
Cases
Expected source and page
16/16
Valid citation identifiers
24/24
Correct abstention
8/8
Dataset
synthetic-regression-v1
Completed cases
24/24
Median / p95 per case
0.847 s / 0.985 s
Evaluation record (UTC)
2026-10-08
Source: the project’s evaluation record, using local inference. Recorded hardware: NVIDIA GeForce RTX 5060 Ti, 595.84, 16311 MiB. Latency includes the full HTTP case and may include initial model loading. These measurements have not been rerun on the machine serving this website.
Evaluated demo
Evaluated demo