Three practices, one thesis: evidence over promises
AI Governance
Risk assessments, evals, and board-ready assurance for regulated sectors — aged care, healthcare, NDIS. Run through GVRN-AI.
AI Automation
Voice agents, workflow automation, RAG systems — every build ships with published evaluation results. Run through GVRN-AI Automation.
Agentic Engineering
Orchestration across the whole stack: cloud data, AI components, agent coordination, approval gates, and the business layer that owns the decisions.
Built and evaluated, not demoed
Clinical safety gates
Healthcare agent with a deterministic emergency-escalation gate wired ahead of the model — 610 passing tests, because "usually right" isn't a clinical standard.
Judge calibration
LLM-judge evaluation pipelines calibrated against human-labelled golden datasets — F1 receipts, not vibes.
Aged-care RAG, cited and evaluated
Production RAG over the Strengthened Aged Care Quality Standards: hybrid retrieval, re-ranking, cited answers, secured API. Runs on three retrieval stacks (local, Azure, BigQuery) under one eval harness. Faithfulness 0.96–1.00, citation accuracy 1.00 on all three, 121 tests — code and evals public.
17 shipped projects
Agents, RAG, evals, dashboards — selected work on the portfolio.
Ask my voice agent anything
Talk to my AI assistant — live now, grounded in my actual work. My own cloned voice is coming; today it speaks with an interim voice. Zero conversation retention either way.
Powered by ElevenLabs Agents · transcripts not stored · you'll always be told you're talking to AI
30 minutes, no pitch
A discovery call about your AI exposure, a build you're weighing up, or a role you think I'd fit. If I'm not the right person, I'll say so and point you at whoever is.