I build things that work for people who need them.
And I break things on purpose to make sure they actually do.
I spent years planning direct actions for Greenpeace — not communications, not fundraising, the Actions unit. Operations against well-resourced opponents who would rather you didn't succeed. That work teaches you one thing above everything else: the optimistic plan is the dangerous plan. Enumerate failure modes before you move.
That thinking didn't leave. It followed me into government cybersecurity, then AI safety — 120 models, 18,000 adversarial prompts, a published methodology — then into everything I build for clients. It is the methodology.
Now I consult, architect, build, automate, and evaluate. Websites, AI systems, security audits, voice agents, agentic pipelines, governance frameworks. Small businesses, regulated industries, anyone with a problem worth solving properly.
Same rigour. Every time.
I built a legal AI system for US federal inmates navigating sentence reductions nobody was explaining to them. I built an operations platform for a local martial arts club — 79 lesson plans, JWT member portal, live kanban, zero build tools. I built a non-coercive AI companion for my own neurodivergent children. I put a hallucination detector on my own CV so fabricated claims can't deploy.
This is not altruism. It is how I'm wired. Do it properly or don't bother.
I'm Autistic and ADHD. I name this because it directly explains the work: hyperfocus means I go to a depth most people can't sustain. Pattern recognition means I notice the failure mode hiding inside the working system. Directness means if your system has a problem, I'll tell you what it is — not a version that's easier to hear.
What I do
The range is broad because most real problems don't fit neatly into one category — and the best solution often crosses several.
Build & deliver
Production systems shipped on time, with no handoff and no disappearing act.
- Websites, web apps, and operations tooling
- Automation pipelines, APIs, and CLI tools
- Multi-site ecosystems with shared brand systems
- Fast, constraint-led, zero bloat
AI integration
LLM systems that work in production — with real safety gates, not vibes-based guardrails.
- LLM pipelines with hallucination detection and CI gates
- Multi-agent orchestration (LangGraph, Anthropic SDK)
- RAG, voice AI, MCP servers, self-hosted inference
- Evaluation frameworks and red-team reporting
Security & evaluation
Adversarial review of systems before an adversary finds what you missed.
- Penetration testing and vulnerability assessment
- Essential Eight and ISO 27001 implementation
- AI red-teaming across models and prompt surfaces
- GenAI governance frameworks for procurement and executives
Where I go deep
Specific domains where I have production experience, not just familiarity.
Healthcare AI
AHPRA-compliant digital presence. Cliniko API integration. NDIS-aware tooling. Clinical decision support systems tested against real case datasets. I understand the regulatory surface and build inside it — not around it.
Legal intelligence
Knowledge graph construction, document analysis pipelines, RAG over large legal corpora. Transparency-first AI reasoning with auditable outputs. Built VERITAS from scratch — a full legal intelligence platform with Neo4j, Cypher, and document extraction.
Government & compliance
Seven years delivering systems integration and cybersecurity for Tasmania's public housing sector. Wrote the department's first Generative AI policy. Essential Eight and ISO 27001 implementation. GenAI governance frameworks that procurement officers and executives can actually act on.
Voice & telephony AI
Real-time AI phone agents: STT-to-LLM-to-TTS pipelines over Twilio with tool use, tri-layer memory, safety oracle, and human handoff. Appointment automation, customer service, crisis detection. Built and deployed end-to-end.
Agentic systems
Multi-agent orchestration for complex autonomous workflows. LangGraph, self-correction loops, Qdrant vector stores, LangFuse observability. Research engines, data pipelines, document processors — systems that do work while you sleep.
AI governance & policy
Model trust scoring, evaluation frameworks, red-team reporting, and governance documentation. Technical findings translated into policy-ready language without losing fidelity. Useful for organisations adopting AI who need to know what they're actually deploying.
Self-hosted & sovereign AI
vLLM inference, local LLM orchestration, data-sovereign RAG. For healthcare providers, legal firms, and government agencies that can't send data to US clouds. The capability without the compliance exposure.
Creative & interactive
AI-powered art installations, generative audio systems, interactive kiosks, and audiovisual experiences. Face-morphing mirrors, real-time music sequencers, granular synthesis tools. Where technical depth meets creative intent.
Here's what that looks like
Evolve Evolution
Four-part ecosystem built for a healthcare practice: chiropractic site, personal coaching brand, author hub, and a private operations dashboard. Shared brand system across all four. AHPRA-compliant content. Password-gated hub for internal docs, kanban, and GitHub issues.
Tanda Pizza
Website for an Italian pizzeria in Lovina, Bali. Five languages — English, Italian, Dutch, Indonesian, Chinese — because that's who walks through the door. Static, fast, no server. WhatsApp for reservations because that's how Bali works. Loads on a patchy connection. Does exactly what it needs to do.
Wolf Clan Zen Do Kai
Three-zone operations platform for a martial arts club. Public site with 24 blog posts and schema.org structured data. Password-gated ops hub with live GitHub Issues kanban and 40+ operational docs. JWT member portal with attendance tracking, belt progression, and 79 lesson plans. Zero build tools — vanilla HTML/CSS/JS on Cloudflare edge.
What clients say
"He has a unique ability to truly listen and understand the essence of someone's work, then translate it into a clear and beautiful digital presence."
"Working with Adrian was exceptional. The process felt thoughtful, collaborative and surprisingly easy, and the result already feels like a true extension of my practice — helping people find solutions to their problems, and answers to their questions."
"I would recommend working with Adrian without hesitation."
How it works
-
Step 1: Talk
Tell me what you're trying to solve. No forms, no pitch deck. I'll tell you honestly if I can help and what I'd do.
-
Step 2: Build
I move fast. Iterative, visible progress. You stay in the loop at every stage. Scope is agreed upfront — no bill shock.
-
Step 3: Stay
Most clients keep me on. Monthly retainer: updates, new features, whatever comes up. I don't hand off and disappear.
Pricing
Project
Fixed scope
2–8 weeks, fixed quote
Scoped, priced, and agreed upfront. No surprises. Right for a defined build with a clear outcome.
Retainer
Monthly
From the month after delivery
Ongoing capacity. Updates, new features, support, whatever comes up. Most clients move here after the initial build.
Consult
Day rate
Half-day or full-day engagements
Strategy, architecture, evaluation, policy. Useful when you need deep expertise for a specific question without a full build.
Rates depend on scope and engagement type. Let's talk.
Have a problem worth solving?
Based in Cygnet, Tasmania. Working remotely across Australia and internationally. Tell me what you're trying to do — I'll tell you how I'd approach it.