Projects

• Evaluation

  rubric-reward-modeling

  LLM-as-judge scorer with a 5-dimension rubric and safety hard gate.

  • 15 passing tests
  • GitHub Actions CI
• Agentic

  agentic-benchmark-harness

  MCP implementation with a deterministic mock tool sandbox and 4-domain YAML task suite.

  • 18 passing tests
  • Safety-boundary failure demo
• Compliance

  continuous-compliance-evidence

  Automated evidence collection and audit-trail generation across frameworks.

  • Multi-framework
  • 100% automated
• Compliance

  compliance-as-code-gating

  Policy enforcement as code, wired into CI/CD with declarative rules.

  • Real-time enforcement
• Monitoring

  drift-monitoring-dynamic-schema

  Real-time drift detection with schema-evolution tracking.

  • Full-stack coverage