Research
The research behind AI you can trust.
We publish what we learn building and testing AI: uncertainty you can act on, evaluation that catches failures before customers do, and the methods we use in client work. Practical, with code where it helps.
Latest · 8 November 2026 · 8 min read
A release gate for AI agents: what we test before every update
AI agents break quietly when a model, prompt or tool changes. Here is the automated gate we put in front of every release: what goes in the test suite, how to set pass thresholds, and how to read zero failures honestly.
- LLM evaluation
- AI agents
- Release gates
- Prompt injection
Talks & publications
Where we’ve presented this work.
PyCon DE & PyData 2025
Talk
Accuracy is not Enough: Building Trustworthy AI with Conformal Prediction
Google Cloud Next
Talk
Speakers on risk-aware AI
ECML-PKDD
Publication
Published research on explainable AI for natural language processing
■ Free 30-minute call
Want this rigour on your AI?
We build and audit AI systems with calibrated uncertainty, stress tests and release gates. Tell us what you’re running and what worries you.