Reliable, sustainable AI operations through effective risk management.
I build open-source tools at the intersection of enterprise operations, AI, and risk. My experience across Amazon, AB InBev, SGS, and chemical manufacturing shapes a practical approach: define the objective, establish accountability, build the controls, and test whether they work.
My thesis: manage risk without sacrificing innovation. These projects explore how people can understand automated decisions, inspect evidence, and retain authority over consequential actions.
Explore the projects · More of the portfolio · Connect on LinkedIn
Compare the same agent request under different control conditions, trace why a decision changes, and form an evidence-backed position. The guided app connects procurement cases, HAZOP analysis, proposed controls, and recorded synthetic experiments.
Research and learning workspace. Browser comparisons illustrate a fixed model; hazards remain unverified and controls proposed. No real purchases or safety certification.
Explore the app · Source · Investigation guide
Self-hosted corporate EHS software for incident investigation, corrective and preventive actions (CAPA), approvals, and evidence across sites. Explore the corporate overview and incident-to-CAPA journey, or run the full application locally.
Application with a synthetic, read-only walkthrough. People retain approval and closure authority; module maturity varies. The demo does not validate a deployment or establish regulatory compliance.
Explore the walkthrough · Source · Self-hosting guide
A Model Context Protocol (MCP) policy gateway and operator console combining Cedar authorization with signed decision receipts. Inspect allow/deny reasons, explore policy scenarios, and verify receipt signatures.
Reference implementation. Public scenarios are synthetic; browser signature checks are real, but do not prove tool execution or issuer trust. Connected operations require the separate development gateway.
Try the demo · Source · Local setup
A governance workspace for following alerts to supporting evidence, recording acknowledgements or escalations, and revisiting review history. The connected development workspace integrates alert and audit services with persistent reviews.
Synthetic public demo and separate connected development application. Demo reviews stay in the browser. Connected mode uses a configured development identity; this is not a production banking or regulatory filing system.
Try the demo · Source · Connected workspace
Investigate order, fulfillment, and inventory discrepancies. Compare faulty and corrected behavior against the same evidence, inspect assertions, and export reports. A Node CLI and library support scenarios and normalized captures; the browser demo has its own evaluator.
Early toolkit with synthetic demonstrations. Missing evidence can produce an inconclusive result. No merchant accounts are connected and demo outcomes do not validate a production integration.
Try the demo · Source · CLI and library guide
| Project | What you can explore | Entry point |
|---|---|---|
| SOC-2 | Local GitHub readiness assessment, finding review, next steps, and export. Three repository checks; not a complete SOC 2 assessment or certification. | Synthetic demo · Local assessment |
| Healthcare Data Exchange | Guided consent, care-team, and regional-access scenarios, with a separate FHIR/OPA/OPAL backend reference. Synthetic data; no clinical use. | Guided demo |
| Asclepius | Fictional visit and referral journeys, linked health records, and a validation lab. Educational FHIR subset; no clinical use or full conformance claim. | Learning app |
| HR-ERP | Multi-tenant HR fixture for agent governance, tenant isolation, and evidence gates. Synthetic employee/HR demo; not a production HRIS or certified payroll product. | Browser demo |
| Professional Service Automation | Consulting realization: approved time, work in progress, draft/issued invoices, and project margin. A scoped reference application. | Source and setup |
| CorpOS | A scripted company day with agent handoffs, tool controls, and human approval. A simulation and reference architecture. | Local demo |
Supporting and historical repositories: healthcare-policy is the OPAL policy mirror; make policy changes in Healthcare Data Exchange. Corporate/Site Harness is an archived reference for digest-bound software-delivery gates and human approval.
- Make the operating problem concrete: a stale approval, an unresolved corrective action, an incomplete shipment, or an access decision.
- Keep evidence inspectable: connect decisions to records, policy, tests, and explicit uncertainty.
- Build usable review workflows: help people understand a result and decide what should happen next.
The repositories span TypeScript/React interfaces, Python tooling, Go services, PostgreSQL, and policy-as-code with OPA/Rego and Cedar. Each project documents its own setup, architecture, checks, and limitations. A passing check supports its stated scope; it does not establish production readiness or certification.
Connect with me on LinkedIn. For project questions, bugs, and contributions, start with that repository's README and contribution guide. For profile corrections, see Contributing; report vulnerabilities through the affected project's security policy.
Portfolio reviewed November 14, 2026. Source and maintenance notes · Profile license: Apache-2.0. Each linked project retains its own license and third-party terms.