← All projects

CURRENT / IN PROGRESS

AI Governance Platform

An open-source platform for governing, evaluating, replaying, and operating AI systems-with reproducibility and operational control treated as first-class concerns.

  • AI systems
  • Governance
  • Control planes
  • Reproducibility

THE PROBLEM

AI execution is difficult to inspect once it leaves a prototype. The important context-inputs, policies, versions, evaluations, and decisions-needs to remain connected as the system operates.

01 / ARCHITECTURE

  • A control-oriented record of execution and its surrounding evidence.
  • Evaluation and replay as operational capabilities, not afterthoughts.
  • Explicit links between changing artifacts and the outcomes they influence.

02 / DESIGN DECISIONS

  • Keep provenance and policy visible in the operational path.
  • Model system behaviour as inspectable, versioned evidence.
  • Favour vendor-neutral boundaries over a tightly coupled platform surface.

03 / CONSTRAINTS

  • Governance has to help teams operate systems rather than merely document them.
  • The system must make failure investigation and reproduction practical.

04 / TRADE-OFFS

  • More explicit records create more structure to manage, but substantially improve explainability.
  • A broad integration boundary gives teams flexibility while demanding careful contract design.

05 / WHAT I LEARNED

  • Operational questions should shape architecture early: what happened, why, and can it be replayed?
  • Useful governance is a system property, expressed through the control plane and its evidence.