v1 · indie AI builder build
AI evals without writing a single test case.
Upload your support transcripts. We auto-generate 8-15 grounded personas, run them against your AI nightly, and diff the results so you know what changed.
Nightly regression — Acme Support Bot
3 personas that passed yesterday FAILED last night.
- "Frustrated First-Time Buyer"0.85 → 0.40
AI said the 50-seat plan is $99/mo. Pricing page says $499/mo.
- "Migrating Customer"0.78 → 0.55
AI never explained CSV import despite 3 prompts.
- "Refund Seeker"0.92 → 0.30
Looped on "please contact support" instead of explaining policy.
The product
Three steps. No test-writing. No rubrics. No bots in your Slack at 11 PM.
01
Upload your real transcripts
CSV from Intercom/Zendesk or paste 20+ utterances. Encrypted at rest. Never trained on.
02
We synthesize 8-15 grounded personas
Every sample utterance is verbatim from YOUR source. Personas are dropped if we can't ground them.
03
Sleep. We're watching.
Nightly runs at 2 AM. Telegram ping at 6 AM only when something that used to work, broke.
Why this exists
The pain we built this for.
- Pain · 01
DeepEval, Confident AI, Promptfoo all want you to write test cases — and you won't, because you have features to ship.
- Pain · 02
Hand-written eval rubrics are a side quest you'll abandon by week three.
- Pain · 03
Your real users are weirder than any rubric you'd think to write.
Pricing
$29/month per AI endpoint monitored.
Persona-driven evaluation with zero setup, grounded in YOUR own data — not generic LLM templates. You've been deferring the eval setup for months. Make it 5 minutes instead of 3 weeks.
Every sample utterance is verbatim from your source. Personas are dropped if we can't ground them. No hand-written rubrics, ever.
- · 1 AI endpoint monitored nightly
- · 8-15 personas synthesized from your data
- · Telegram alerts only when regressions appear
- · 14-day pass-rate sparkline
- · Claude Haiku 4.5 judge
First audit free · no card required