Architecture
Inspect service boundaries, data flow, trust boundaries, dependency choices, and separation of concerns.
Engineering Proof Lab
Inspect the systems behind the portfolio.
Public reference implementations for healthcare and regulated AI. Review the architecture, control boundaries, tests, evaluation logic, audit evidence, and operational failure handling directly in the source.
12Public proof repos
11Implementation areas
3Core languages
4Flagship systems
What to Evaluate
Read these repositories as architecture samples. The strongest signals are not feature count; they are explicit boundaries, enforceable controls, verifiable behavior, and operational discipline.
Controls
Trace where identity, PHI handling, policy enforcement, approvals, and audit evidence enter the system.
Verification
Review test coverage, evaluation criteria, deterministic checks, failure cases, and evidence artifacts.
Operations
Evaluate observability, retries, cancellation, backpressure, deployment gates, and incident traceability.
Engineering Position
These repositories contain no client code or protected data. They isolate reusable patterns for the engineering decisions that matter in production regulated AI.
Execution evidence should be inspectable, signed, and portable.
Regulated AI needs policy gates at tool boundaries, not only prompt guidelines.
Healthcare AI systems need PHI controls, audit logging, and evaluation gates before deployment.
Production agents need observability, backpressure, cancellation, and regression tests.
Flagship Systems
Four systems carry the primary architecture signal: governed retrieval, model risk, validated delivery, and verifiable execution evidence.
healthcare-rag-platform
Healthcare RAG reference architecture with PHI detection, retrieval guardrails, audit logging, and evaluation metrics.
Public reference implementation
10 tests passing
model-governance-framework
Model-governance controls for healthcare ML, including bias analysis, fairness metrics, model cards, and approval evidence.
Public control framework
18 tests passing
mlops-healthcare-platform
Regulated MLOps reference architecture with IQ/OQ/PQ validation patterns, MLflow integration, registry controls, and audit trails.
Public platform pattern
12 tests passing
agentic-receipts
Cryptographic receipts for agent traces using hash chains, Ed25519 signatures, and JCS canonicalization.
Public verification primitive
26 conformance vectors passing
Supporting Components
These repositories isolate integration, runtime, policy, evaluation, and application concerns used by the flagship architecture.
Healthcare RAGAI GovernanceRegulated MLOpsCompliance AutomationHealthcare InteropAgent EvidencePolicy ControlsEvaluationEvidence UXHealthcare AgentRuntime Reliability
compliance-automation-suite
HIPAA references, PHI validation, security controls, and traceability.
Tested component
2 tests passing
fhir-integration-service
FHIR R4 resource handling, integration boundaries, and exchange patterns.
Tested component
4 tests passing
agentic-trace-cli
Receipt canonicalization, signature verification, and CLI ergonomics.
Verification primitive
2 tests passing
agentic-policy-engine
Deny-by-default policy gates, tool-boundary decisions, and receipt output.
Tested component
7 tests passing
agentic-eval-harness
Scenario gates, bypass resistance, deterministic behavior, and regression tests.
Tested component
41 tests + 8 scenarios passing
agentic-evidence-viewer
Trace-chain rendering, redaction handling, and evidence review UX.
Application scaffold
4 tests, type check, and build passing
agentic-member-assistant
Identity-scoped retrieval, safe explanations, and healthcare agent flow.
Application scaffold
38 tests and build passing
agentic-streaming-backend
SSE streaming, backpressure, cancellation, retry, and observability.
Tested component
51 tests and build passing
Architecture Review Path
Begin with the integrated architecture and its trade-offs, then inspect the four flagship repositories. Use supporting components to trace specific controls and failure behavior.