2,464 passing tests across four independent governance implementations, built solo. Measured 2026-09-20. One of the four is public and documented end to end in the Governed Code-Change Evidence package, including three adversarial checks run against my own confinement and a disclosure of the run where that confinement was switched off. The other three are private and available for walkthrough.
Applied AI Systems Architect. I design AI-native workflows, agent systems, evaluation layers, and deterministic controls that turn business processes into governable, measurable software.
Governance is my differentiator, not my whole category: I build the enforcement layers, audit trails, and authority separation that keep autonomous systems from doing irreversible damage — and I build the products and integrations that sit on top of them.
Start here: Governed Code-Change Evidence (evidence package) → Spine Lite (public governance kernel) → Governed Swarm (reference architecture) → StemForge (applied engineering).
Repo access & maturity legend: 🟢 public · 🔒 private (case study available during interview). Maturity tags:
reference·prototype·pilot·production-candidate. Where a repo is private, follow the public substitute or ask for a walkthrough.
Executed test suites, not estimates. Dated so drift is visible rather than hidden.
| System | Access | Passing | Test:source LOC | Measured |
|---|---|---|---|---|
| Commerce Authority Workbench | 🔒 | 1,377 | 2.1 : 1 | 2026-09-20 |
| M87 Crew | 🔒 | 551 | 1.4 : 1 | 2026-09-20 |
| Spine Pro | 🔒 | 323 | 1.5 : 1 | 2026-09-20 |
| M87 Agent Architect | 🔒 | 213 | 0.5 : 1 | 2026-09-20 |
| Total | 2,464 | 1.08 : 1 |
Test code exceeds source code across the portfolio. Spine Pro is reported at 253 passing in the July evidence package and at 323 today; the suite grew, and both numbers are kept rather than quietly replaced.
Commerce Authority's suite includes a test that runs git ls-files *.md and fails the build when the repository's own documentation overclaims.
M87 Studio — a governance-first execution stack for AI agents, plus the applied products built on it.
| Layer | Project | Access | Maturity / License |
|---|---|---|---|
| Governance kernel | Spine Lite (Python · JVM · .NET) | 🟢 | MIT, multi-runtime, alpha |
| Control plane | Spine Pro | 🔒 | BSL 1.1, gate / token / executor / receipt ledger |
| Reference impl | Governance Sandbox | 🟢 | BSL 1.1, reference |
| Executable spec (BPMN) | m87-governed-loop | 🟢 | MIT, Python + Java, reference |
| Agentic substrate | Governed Swarm | 🟢 | BSL 1.1, prototype |
| Audit | m87-audit-agent | 🟢 | MIT, prototype |
| Evaluation | m87-retrieval-evals | 🟢 | Retrieval + generation metrics, fail-closed judge, CI gate |
| Bridge | m87-governed-bridge | 🟢 | Claude ↔ Gemini, archived research |
| Applied system | StemForge | 🟢 | MIT, local GPU audio workstation |
| Live product | Resonance | 🌐 live · source private | Hosted product, live Stripe billing, sole on-call operator |
- Proposal ≠ Execution. Agents propose actions. Governance executes them.
- Authority separation. Decision logic and execution logic stay on different sides of a contract.
- Fail-closed by default. Ambiguity halts. No silent recovery.
- Artifact-backed completion. No work is "done" without verifiable receipts.
- Model interchangeability. No design depends on a single model.
These invariants guide the M87 architecture. Spine Lite provides the deterministic policy-and-effects kernel at its foundation. The implementations vary by workflow; the control principles remain constant.
- Applied AI roles — discovering workflows, building products, designing integrations, and directing AI-native implementation.
- Advisory — governance architecture for AI-native teams shipping autonomous systems.
- Design partners — for Spine Pro in regulated and high-assurance workflows.
📍 Portland, OR · 🌐 m87studio.net · 💼 LinkedIn