Skip to content

Gabriel Goodhart

AI Governance & Applied AI
Oakland, CA youdontneedmy.help Contact via /submit
38
AI systems built
25
live in production
2
in the care sector
Jul 2023
building with AI since
Summary

AI governance and applied AI leader with 25 years leading high-stakes work in regulated industries and three years building production LLM systems. I have built 38 AI systems since July 2023, put 25 into production, and designed two care-sector products around consent, privacy, auditability and human oversight — PHI handling under HIPAA as a first-commit design constraint, not a retrofit. I build the review, evaluation and approval workflows that tell a team what can ship, what needs work, and what must stop — and I turn technical risk into practical decisions for executives, operators and delivery teams. Outright failures alert me; drift I still catch by inspection, and I say that out loud rather than claim monitoring I cannot show. Every demo linked below runs on fictional sample data; no client or personal data is exposed anywhere.

Experience Across Regulated Operations

The AI work on this page sits on a 25-year operating career in regulated industries. Organizations, titles and dates below; references on request.

OOG Consulting, Oakland, CA
2014 – Present
President and Senior Project Manager
AI systems and AI governance practice · 2023 – Present
  • Founded Stillwell, a family reassurance and answering line for older adults, and wrote its formal governance posture: consent that fails closed with teach-back and revocation, voice biometrics treated as encrypted deletable data, no retained call audio or verbatim transcripts, honest AI identity disclosure, and a fixed family-alert escalation ladder.
  • Operate that governance the way health systems run theirs: a posture document with an explicit authority order, a state-by-state counsel review matrix (California, Colorado, Maine), a dated implementation-vs-policy divergence audit, and a 3,200-test release gate that fails builds on policy violations.
  • Founded Calm Couch, a private-beta practice platform for couples counselors and their clients: no free-text clinical notes or conversation transcripts stored, synthetic data only until compliance approval, row-level security isolation, append-only consent and audit evidence, and legal review plus BAAs gating any real-client use.
  • Built 38 production AI/LLM systems (25 live) with governance controls as first-class features — the systems, evaluation cascades, guardrails and drift instrumentation this page documents — and trained non-technical business owners in safe AI adoption with plain-language SOPs and risk guidance.
Regulated-industry risk and compliance programs · 2014 – 2022
  • Directed remediation of state-licensed commercial agriculture operations (250 to 3,000 acres) across multiple jurisdictions; delivered 13% under budget.
  • Wrote risk mitigation and compliance plans covering endangered-species habitat, cultural-artifact sites, watershed rules and state licensing requirements.
  • Convened and aligned attorneys, compliance experts, licensing consultants, biologists, hydrologists, engineers, contractors and operations teams — 10-plus-person multidisciplinary crews — and ran full-scope audits across financials, security, staffing, safety and SOPs.
  • Built real-time KPI and risk dashboards for live project auditing; a behavior-based safety program for field staff cut workers-comp claims 70% in year one.
Dosa Distributions LLC, Oakland, CA
2007 – 2014
Managing Director
  • Designed and brought to market fertigation technology now standard for nearly 40% of hydroponic growers; the partnership with Dosatron led to acquisition.
  • Grew enterprise sales 328% in three years; negotiated 10+ vendor contracts.
Ameriprise Financial, San Francisco, CA
2001 – 2007
Corporate and High-Net-Worth Financial Planner
  • Advised corporate and high-net-worth clients in a FINRA/SEC-regulated practice under strict suitability, privacy and documentation requirements.
  • Held Series 7, 63 and 65 registrations plus California Life and Health licenses; exceeded quota 241% (2002) and ranked top 10% nationally for relationship management.
Education & Credentials
Selected Work
TradeTEST.TRAINING
AILIVE
Paying customers study for a state licence exam in English or Spanish, and no translated question ships until five checks agree it kept its meaning (~3,400 items through the gate, resumable, cost-capped). Billing, spaced-repetition scheduling and content pipeline all shipped end to end.
Groq · Next.js · Supabase · TypeScript · Stripe · SM-2 algorithm
▶ Read the case writeup
Sea Star — Groq Publish Automation
AILIVE
Non-technical staff publish to four channels from one post, nightly, with no engineer in the loop. A single model call fans out to Facebook, Instagram and email formats; it runs as an admin tool in production.
Groq · Next.js · social publish API · nightly cron
▶ Run the demo
Multi-Model QA Cascade
AI
No model grades its own homework: three providers propose in parallel, plain rules pick the winner, and a person approves before anything is committed to the database. $0 to run locally, which is why it actually gets used.
Ollama · Groq · OpenAI/Codex · deterministic scoring · human-in-loop · SQL
▶ Read the case writeup
agent-gate — Safety Fence
AI
Stops a coding agent from leaking secrets, deleting data, or claiming success without proof. Any edge it does not recognise halts the run rather than guessing. 53 tests, zero dependencies, built as a standalone reusable safety layer.
deterministic rules · fail-closed · 53-test suite
▶ Read the case writeup
Vision Factory — Dual-Model Consensus
AI
Stops an inconsistent asset run before it reaches production: one model generates, a second verifies placement, and the run halts when they disagree past a fixed threshold. A status board keeps a long run inspectable while it is still cheap to stop.
OpenAI GPT-4o · Gemini vision · consensus gate · asset pipeline
▶ Run the demo
Find Your Vote
LIVE
Every voter sees exactly why a candidate ranked where they did, and the same address always gets the same answer — audited for that, run after run, across 25+ public-record sources.
Next.js · TypeScript · determinism audit · 25+ public record sources
▶ Run the demo
blood-suga — Vision Meal Analysis
AIDEMO
Shows you how wrong it is, live, while you use it: every estimate is scored on screen against USDA-derived ground truth, and the offline fallback says on its face that it is a fallback. Production path uses vision models; the demo scores a live text call.
Groq · llama-3.3-70b · dataset-backed eval · MAPE · offline fallback
▶ Run the demo
Capabilities & Stack
Models orchestrated
Claude OpenAI / Codex / GPT-4o Groq Gemini vision Local Ollama Stable Diffusion
Infrastructure
Next.js TypeScript Supabase Realtime Playwright Tailwind Vercel
AI / Pipeline
Deterministic pipelines Dataset-backed evals Non-LLM scoring Multi-model routing Fail-closed safety MAPE / eval loops
Systems shipped
SaaS (Stripe + SM-2) Publish automation Human-in-loop workbench Agentic safety fence Schema-driven intake Unit-economics engine
How I Work
Deterministic first. LLM output is wrapped in rules, scoring, and gates so the system behaves predictably even when the model does not.
Verifiable by design. Evals run against real labeled data; MAPE and pass/fail rates are visible in the demo, not buried in a notebook.
Fail-closed safety. Any uncaught edge halts the pipeline. No silent fallback to "probably fine." agent-gate is a dedicated reusable layer for this.
Human-in-the-loop. Operator review gates are a first-class component, not an afterthought. audit-kit ships a typed JSON workbench the next stage reads back.
No client data in demos. Every demo tile on the portfolio runs on sample data only. Real products operate under separate, isolated data paths.