Customers
/
Veda Health

Passed a clinical safety reviewwith evals.

Veda Health scores every model change against 2,400 graded cases before it reaches clinicians.

Passed a clinical safety review with evals as the evidence.
2,400
graded cases per deploy
0
regressions shipped
EU
data residency
14 days
to safety approval

The challenge

Veda Health summarises patient intake notes for clinicians. Their safety board would not approve a new model without proof that it was at least as good as the old one on real cases.

The change

The team built a versioned test set of 2,400 cases from redacted production traffic, graded by clinical rubrics. Every prompt or model change now runs through the Lattis CI gate, with inference pinned to the EU region.

evals: suite: clinical-intake-v4 cases: 2400 region: eu-west gate: score >= baseline

The review board wanted numbers. Evals gave us numbers.

Dr. Priya Raman
,
Clinical AI Lead
,
Veda Health

The result

The review board approved the switch with the eval report as evidence. Since then, no change that lowered the score has reached production.

Before Lattis, and after.

Measured over 90 days
Review time per model change
Before
6 weeks
After
2 days
Manual review hours per change
Before
60 h
After
4 h
Regressions reaching clinicians
Before
3
After
0
Closing the gap

The platform underneath your AI product.

Start with 1M free tokens. No card required.