Clarity for high-stakes ML decisions.
From the first experiment to the final call.
tessel is decision infrastructure.
decision = f (signals | context)
It’s the layer your team makes decisions on: a record of every decision, the right signals in front of whoever makes it, and faster decisions as the evidence builds, until routine ones run themselves.
See how it worksRecord
Every decision, what it was based on, and what happened next.
Understand
What your model really relies on, and where it isn’t safe to trust.
Automate
Autonomy you can justify: trusted only as far as the evidence goes.
Our point of view
Start from the decision, not the metric.
Most evals and benchmarks are built before anyone knows the decision they’ll inform. We work back from the decision: only the questions that could change it, each asked so the answer holds up.
The answers appear here as each question is asked.
For teams that can’t afford to guess.
From the first model to the submission, and every update after.
Built for regulated work: held-out set discipline, evidence for a 510(k) or De Novo submission, and predetermined change control plans.
Explore healthcare
Admitted with community-acquired pneumonia and treated with IV antibiotics. Started on metformin 500 mg twice daily. History of atrial fibrillation. Follow up with primary care in two weeks.
- ✓Pneumonia, IV antibiotics: in the admission notes
- ✓Metformin 500 mg twice daily: in the medication record
- ✕Atrial fibrillation: not in the record. Flagged for a clinician.
Field notes.
All postsA model is built one decision at a time
Why high-stakes ML needs decision infrastructure, and what we’re building at tessel.
Coming soonThe decision benchmark, v1
How far you can trust a model or agent with a high-stakes decision, tested on real decisions.
Coming soon