Gemvis
Gemassay
Sign in
Gemvis quality, measured

Know how good Gemvis is, every day

Gemassay checks the mine chatbot's answers around the clock. When quality slips, it shows up here first, before anyone has to go looking for it.

Sign in to Gemassay See how it works
0
Evaluation layers
Validators, RAGAS, ASSAY, Harness
0x
Checks a day
Validators run every three hours
0+
Benchmark questions
Golden set across eight topics
How it works

Four layers of evaluation

Each layer looks at a different side of quality, from a quick check on every answer to one weighted score. Pick a layer to see what it covers.

Validators

Light checks that run eight times a day. They catch stale dates, missing figures, and answers that come back empty or refuse to help.

Sign in to see the results
  1. 1FreshnessDates and periods in the answer are still current
  2. 2FormatRequired keywords, figures, and percentages are there
  3. 3Non-answerEmpty replies, refusals, and errors get flagged

What Gemassay grades

Six quality dimensions, measured on every Gemvis answer.

FidelityContent is right
FreshnessData is current
CoverageAnswer is complete
RoutingRight agent picked
FaithfulnessGrounded in data
CorrectnessMatches the key

On a schedule

Every 3 hours
Validators

Eight times a day: dates, formats, and empty answers.

Daily 02:00
RAGAS

Faithfulness, correctness, and relevancy, judged by an LLM.

Daily 04:00
I/O mode

Four combinations of text and voice, in and out.

Ready to see how Gemvis is doing?

Sign in for the dashboard, run history, and per-dimension reports.

Sign in to Gemassay