MW. Résumé / CV

Agent build case study

Building a governed AI agent on an open knowledge base

First live cloud run 27 September 2026.

I designed and built a custom AI agent that runs in the cloud, draws everything it knows from an open knowledge base, and can only act inside policy written as code. I tested it on a live, high-stakes domain: my own job search.

In its first nine days the system tracked 234 roles and logged 8,782 deterministic decisions at zero model cost. All 208 automated tests pass. It runs entirely in the cloud: a person presses Run in a web console, and the agent works inside rules it cannot change.

What it demonstrates

  • Agent design. One bounded agent step inside deterministic code, with a tool allowlist and no credentials of its own.
  • Knowledge engineering. An OKF v0.2 knowledge base that the agent, the code and people all read from one record.
  • AI governance. Policy as code, human holds, a truthfulness gate on every claim, and Git history as the audit trail.
  • Cloud security. Zero-trust access, a least-privilege token, encryption at rest and controls against prompt injection.
  • Cost engineering. A tiered model cascade that keeps most decisions at zero tokens.
  • Delivery. Serverless control plane, CI/CD on GitHub Actions and 208 automated tests.

Figures come from the project's own repository, test run and run ledger as of 27 September 2026.

More projects