Principal Technical Program Manager, AI Delivery
Principal Technical Program Manager with 17+ years leading enterprise platforms, AI programs, and complex technology delivery. Builds the operating models, governance, and release controls that move programs from executive intent through production. Led portfolios up to $51M, scaled organizations to 60+ people, and launched cloud and AI platforms for enterprise users.
Representative outcomes:
- $51M AI and data portfolio delivered across 13 regulated brands, with the delivery organization scaled from five people to more than 60
- Enterprise RAG platform taken into governed production for 500+ users: source governance, retrieval, evaluation, access controls, and a named operational owner
- Diagnostics data ecosystem modernized across ~27,000 instruments — retrieval efficiency up 65%, a delivery cycle cut from roughly six months to three weeks
- Content operations redesigned to shorten review cycles 35% and raise approved-content reuse 2.3×, with human and MLR approval unchanged
The repositories below are reference implementations: the governance and evidence patterns those programs relied on, written as working code rather than described in a deck.
Six interoperating components for verifiable agent execution. The meta-repo agentic-evidence (v0.1.0) carries the suite-level reference architecture, standards interop, glossary, roadmap, and versioning policy. The six components below implement it.
Spec ──→ Tooling ──→ Governance ──→ Evaluation ──→ Review
| Layer | Repository | Role |
|---|---|---|
| Spec | agentic-receipts | Receipt schemas, canonicalization, hash-chain vectors |
| Tooling | agentic-trace-cli | init → append → sign → verify → redact → export |
| Governance | agentic-policy-engine | Deny-by-default evaluation with decision receipts |
| Evaluation | agentic-eval-harness | PHI redaction, budget caps, bypass attempts |
| Review | agentic-evidence-viewer | Client-side bundle verification |
| Provenance | agentic-artifacts | Artifact manifests with integrity hashing |
Architectural invariants:
- Hash-chained receipts (SHA-256) make tampering, reordering, and omission detectable
- Every allow/deny emits a decision receipt — auditors see exactly why
- Ed25519 signatures for non-repudiation across trust boundaries
- Redaction preserves integrity verification
| Repository | Domain |
|---|---|
| healthcare-rag-platform | HIPAA RAG: PHI detection, guardrails, audit logging |
| mlops-healthcare-platform | FDA 21 CFR Part 11 model validation (IQ/OQ/PQ) |
| enterprise-llm-integration | Secure LLM adapter patterns with governance |
| model-governance-framework | Bias/fairness, FDA-ready model cards |
| clinical-nlp-pipeline | Medical NER, UMLS linking, ICD-10 coding |
Verifiability — receipts, canonicalization, hash chains, signatures Governance — deny-by-default policy, decision receipts, bypass detection Evaluation — scenario harnesses, red-team cases, regression gates Regulated AI — HIPAA Safe Harbor, FDA 21 CFR Part 11, UMLS, ICD-10, HL7 FHIR Stack — Python · Rust · TypeScript · FastAPI · Kubernetes · Azure ML · AWS SageMaker
Open to: Principal Technical Program Manager and AI delivery leadership roles where platform execution, governance, and organizational scale matter.