Skip to content
View skirby359's full-sized avatar

Block or report skirby359

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
skirby359/README.md

Steve Kirby, J.D.

A technology attorney who builds the AI he governs — and who ran the enterprise systems AI now sits on top of. Most AI-governance people have one of those legs; I have the triangle. I design and ship secure, citation-grounded LLM systems and own the legal, privacy, and risk frameworks that keep them defensible. Spokane, WA · open to remote / relocation / international.

  • ⚖️ Practicing attorney (WA Bar, 2010) — technology, IP, privacy (HIPAA/GDPR), cyberlaw
  • 🏛️ AI governance — NIST AI RMF, EU AI Act, model risk & audit; IAPP AIGP (certified 2026, No. 192392076)
  • 🛠️ Hands-on builder — local-first RAG, anti-hallucination verification, evaluation harnesses
  • 📊 Enterprise data & cloud architect (Nike, Costco) — the systems AI now runs on
  • 🎤 Authored & taught a WSBA-accredited CLE on the ethical use of LLMs in law (2025)

🔭 Featured work — assurance results, not demos

wa-legal-ai-showcase — Air-gapped legal-research AI for Washington State law (public showcase of the architecture; the production system, corpus, and prompts are private). A 7-stage retrieval pipeline (rewrite → retrieve → rerank → pack → answer → verify → render) over 428K legal authorities that refuses to hallucinate: every citation is verified against the corpus before it reaches the user. On a 150-question attorney-reviewed gold set it scores 98%+ citation accuracy with zero hallucinations, fully on-prem — the kind of control a Head-of-AI-Risk has to specify and be able to verify. Python · FastAPI · PostgreSQL/pgvector · Ollama · Claude · Docker

wa-cite-check — Catches the mistake getting lawyers sanctioned. Point it at a motion (.docx/.pdf) and it flags every fabricated, mis-named, or wrong-year citation, plus authorities that have been overruled or repealed — fully offline. An optional LLM judge checks whether each cited authority actually supports the proposition it's cited for. Python · SQLite · click · LLM-as-judge

muni-bond-ai-showcase — The same anti-hallucination architecture, re-pointed at federal tax-exempt municipal bond law (IRC, Treasury Regs, IRS guidance, Tax Court). On a 12-question attorney-authored gold set: zero invented citations, 98.9% claim-to-evidence coverage, 100% out-of-domain refusal — proof the methodology is a reusable platform across regulated domains, not a one-off. Python · FastAPI · PostgreSQL/pgvector · local LLMs (vLLM/llama.cpp) · Claude

complyguard-showcase — Compliance-by-design, automated. Crawls an e-commerce site and audits it against state + federal consumer-protection law across all 52 US jurisdictions — a 71-rule engine plus a Claude analysis pipeline that returns risk scores, statutory citations, and plain-English fixes. A full-stack product, not a demo: authentication, Stripe billing, background jobs, PDF/DOCX reports. Next.js · TypeScript · PostgreSQL/Prisma · Stripe · Playwright · Claude


🧰 Tech I build with

Python · Claude · Ollama (Llama / Mistral / DeepSeek / Qwen) · pgvector · FastAPI · Docker · AWS & on-prem/local · RAG · evaluation & guardrails · model fine-tuning

🤝 What I'm looking for

AI governance sits where legal risk, enterprise systems, and real deployment meet — I'm not changing careers, these threads converged. I'm looking to own an organization's AI-governance program end to end as an operating leader (AI governance, responsible-AI engineering, or trustworthy AI in a regulated domain), and I take on fractional / interim Chief-AI-Risk engagements.

📫 kirby@tikorconsulting.com · LinkedIn · Tikor Consulting

Pinned Loading

  1. complyguard-showcase complyguard-showcase Public

    AI-assisted consumer-protection compliance scanner for e-commerce — Next.js 16, Prisma, Stripe, and a Claude Haiku/Sonnet analysis pipeline over 71 rules across 52 US jurisdictions.

    TypeScript

  2. email-metadata-analytics email-metadata-analytics Public

    End-to-end analytics for organizational email metadata (Date, Size, From, To) — network analysis, community detection, and timing insights from header data only. No message content accessed.

    Python

  3. skirby359 skirby359 Public

    Personal Description

  4. wa-cite-check wa-cite-check Public

    Offline citation auditor for legal briefs: flags citations that don't exist, misstate the case, or point to overruled authority — the failure mode drawing court sanctions.

  5. wa-legal-ai-showcase wa-legal-ai-showcase Public

    Python