Assistants grounded in content
Conversational assistants that answer from our educational content, framed by the learner’s level and curriculum.
Artificial intelligence
Useful to the learner, or not at all
We use language models where they cut real effort — understanding a question, producing an exercise, correcting a mistake — and we decline decorative uses.
What we operate
Four uses in production
Agent platform
An internal foundation to define, version, evaluate and run agents: one framework for every team, rather than scattered integrations.
Adaptive learning paths
Progress, mistakes and time spent feed spaced repetition and the selection of the next exercises.
Production and quality control
Assistance with writing and reviewing content, and automated detection of defects in applications before release.
Our principles
The limits we set ourselves
People decide
No educational content is published without human review. A grade or a correction produced automatically remains a proposal.
Grounded answers
Assistants rely on our verified content and state their limits rather than inventing a plausible answer.
Data under control
Learner data is not used to train third-party models. Processing is documented in our privacy policy.
Measure before rollout
A use is generalised only if it improves an observable outcome: understanding, success, time saved.
An AI use case to qualify?
We help our partners separate a genuine gain from an announcement effect.