The platform we tried to break ourselves
- Client
- An online education platform
- Service
- AI backend and safety
- <0.1s
- response time, guardrails on
What was broken
They were building a platform where an AI answers learners' questions directly. In their subject area, a single confidently wrong or over-personalised answer is not an embarrassment, it is a liability. They needed the backend built and the confidence that it would hold up.
What we built
Two things, together. First, the platform backend, the working system behind the product. Second, the AI governance and compliance layer: rules for what the AI may answer, what it must refuse, and how those limits are enforced in the code, mapped to recognised governance frameworks rather than invented in-house. Then we tested it like an attacker would, with adversarial testing and jailbreak attempts against our own build, closing every hole we could find until it stopped breaking.
What happened
They launched knowing exactly what their AI will and will not say, with documentation to prove it, not a hope that nobody probes it.
Fifteen minutes.
Bring what you sell and where your leads come from now. We'll tell you what we'd build and what it would cost.
Fifteen minutes. If we can't help, we'll tell you who can.