Materiality Data Ltd

Applied AI engineering, with the evidence to go with it.

Agent and voice systems. The evaluation, audit and runtime controls that show what they actually do.

Work
  • auditk

    A scoring protocol for agent traces: an open specification, a reference implementation and a technical report.

    Technical reportGitHub

  • bossyk

    A reference monitor for tool-using agents: govern the effect, attest the record. Public sandbox.

    GitHub

  • stt-bakeoff

    A vendor-neutral harness for measuring speech-to-text engines on your own audio.

    GitHub

  • constellaration

    Stellarator optimisation on Proxima Fusion's open fusion benchmark, with results on the public leaderboard.

    Benchmark

  • Writing

    Notes on audit trails, agent governance and evaluation.

    Substack