Matthan Bird 的 AI 作品集：agent-governance-kit 与 value-registry 等治理工具

Matthan Bird, JD — AI Portfolio

AI strategy and solutions leader with end-to-end ownership of enterprise AI transformation — opportunity discovery, business-case development, technical architecture, production delivery, governance, and value realization. Currently PMC Lead Data Scientist, Enterprise Solution Architect & Decision Intelligence Consultant at Duke Energy; previously Lead Experimental AI Scientist in Duke Energy's Emerging Technologies R&D lab, and before that led Conversational AI at Spectrum (Charter Communications) at 18M+ monthly interactions. J.D.; U.S. Navy submarine electronics & intelligence veteran.

The through-line of this portfolio is governed AI: systems that ship with their audit trail, their evidence standards, and their human-accountability structures built in — not bolted on. Both installable packages are CI-tested with enforced coverage floors, the value-registry engine refuses to report a figure without declared provenance, and every dataset here is synthetic.

Capabilities

Agent Governance — the flagship

packages/agent-governance-kit/ — an installable, zero-dependency Python package of governance primitives for agentic systems: a hash-chained tamper-evident audit log, redaction and policy-as-code check stages, a three-outcome failure state machine whose startover path is human-only and verified against a persisted approval boundary, and a cross-process human-in-the-loop gate. 60 tests, mypy strict, demo in under a second. agentic-ai/ demonstrates the architectures it governs — LangGraph RAG, MCP server, multi-agent orchestration — and ai-governance/ holds the policy and strategy documentation layer.

Value Engineering & Decision Intelligence

packages/value-registry/ — an AI opportunity value model and model-risk registry sharing one data model: the evidence-classed figure. Every number declares how it was known (documented / estimated / modeled) and with what confidence; the engine refuses unclassified figures, aggregates inherit their weakest input's class, and lifecycle gates can forbid scaling on modeled money. ai-solution-lifecycle/ is the full-stack expression: a FastAPI + Vue 3 platform for evaluating and managing AI deployments — value scoring, risk registers, RACI, SLAs, and LLM-backed document generation.

Conversational & Speech AI

conversational-ai/ — the full lifecycle of a production voice channel: confidence-based IVR intent routing with LLM fallback, enterprise prompt engineering with function-calling schemas, NLP analytics dashboards, and an ASR optimization lab that batch-tests speech recognition configurations against background-noise scenarios across Azure Speech, Deepgram, and Speechmatics.

Critical-Infrastructure Defense

infrastructure-defense/ — defensive tooling built around the constraints of regulated infrastructure environments: file-integrity monitoring, network inventory and exposure auditing, deception-based early warning, envelope encryption with key rotation, and TLS compliance verification. Passive-first, cloud-free, auditable by default.

Vision, Robotics & Autonomy

computer-vision/ — YOLOv8 + BLIP-2 monitoring with time-series anomaly detection and multi-level alerting. robotics/ — mission orchestration for Boston Dynamics Spot: checkpoint-based inspection scheduling with post-mission reporting.

Data Foundations

digital-twin/ — privacy-first multi-database architecture: encrypted vector search, knowledge-graph analysis, and metadata lineage tracking with sensitivity-classified encryption.

Getting Started

git clone https://github.com/XenoThanBird/Portfolio.git && cd Portfolio
python -m venv .venv && source .venv/bin/activate   # .venv\Scripts\activate on Windows
pip install -r requirements.txt

# Fastest end-to-end demos (each < 60s):
pip install -e packages/agent-governance-kit && python packages/agent-governance-kit/examples/demo.py
pip install -e packages/value-registry && cd packages/value-registry && make demo

Each section's README carries its own file guide, runnable examples, and module-specific requirements. Python 3.9+. The CI workflow runs a 3.9/3.11 matrix with compile checks, package test suites, lint, and secret scanning; GitHub code scanning (CodeQL, repository default setup) runs alongside it.

License & Contact

Licensed under Apache-2.0 — use it, learn from it, build on it. No real operational data, credentials, or employer-internal content appears anywhere in this repository; every dataset is synthetic.

LinkedIn: linkedin.com/in/matthan-bird-jd-mdb28173
Email: bird.matthan@gmail.com

"The future belongs to those who design it." — Matthan Bird