Product / Quality and insights
Measure your search, don’t assume it.
In short
troveGEN includes an evaluation harness that writes test questions from your own documents, scores search and answers, and compares two settings side by side. Built-in dashboards show usage, latency, index health and every API key.
Evaluate
Questions from your documents
Generate up to 200 questions per run. Each is checked against its source, and unanswerable ones are verified to be unanswerable.
Real scores
Hit rate, rank quality, answer correctness, support by sources, refusal accuracy and answer completeness, plus latency percentiles.
A/B any setting
Run the same questions with different settings, such as ranking, expansion or filters, and see per-question wins and regressions.
A safety gate for model changes
A cheaper model can be switched on only after evaluation shows it holds quality.
Observe
Overview and analytics
Requests, outcomes, latency by stage, grounding and tokens, with a table view for every chart.
Knowledge base and index pages
Documents, status, chunk inspector, vector index details and re-embedding in one place.
API key page
Create, scope and revoke keys, and see who made each.
Data health
Spot tiny or oversized passages, missing structure, low self-containment and open conflicts between documents.
Watch ingestion live
An observatory view shows a file moving through each stage as it happens.
Privacy of the numbers
Counts, not content
The query log stores outcomes, timings and sizes, never the question or passage text.
How to prove a change helps
Generate
Create a question set from your documents.
Baseline
Run it with today’s settings.
Change
Run it again with the new setting.
Compare
Keep the change only if the scores and the regressions support it.
Questions people ask
How do I measure RAG quality?
Use a set of questions with known answers from your own documents, then score retrieval (did the right passage come back, and how high) and answers (correct, supported, declined when unanswerable). troveGEN generates and scores such sets for you.
Can I compare two configurations?
Yes. Run the same questions with two settings and compare scores, with the questions that improved or regressed listed.
Does the dashboard store my questions?
No. The query log records outcomes, timings and sizes only.
Last updated 2026-10-26. Figures are from troveGEN’s own tests on the dates given; ask us for the method.
Prove it on your own documents.
Start free and generate your first question set.