Auckland / AI implementation, evaluation & governance
Confidence
is not
accuracy
I help businesses get ready for AI, and I build the systems that get them there.
Nine years automating and redesigning how work actually gets done at Danone, Tourism Holdings and Dell, and three AI systems now in production that I designed, built and run. Most transformation work stops at the strategy deck. I take it through to something people use on a Monday morning, with the value case and the success measure attached.
Turning information into a business ready for what comes next
Every organisation is sitting on more information, and more good ideas, than it can act on. The work is finding the patterns worth building on, turning them into automation and AI that people genuinely trust, and leaving the business more capable than it was. The same eight stages run every time, with safety designed in from the first one.
Drag to turn · hover a node
Process discovery before technology. Where the effort actually goes, described by the people doing the work rather than by an org chart.
Three systems, built and running
Each one has its own case file: what the business needed, what I built, how it was tested, and what it delivered.
KiwiStart
An AI assistant for people arriving in New Zealand, answering questions about visas, banking, tenancy and study. I designed it, built it and run it on my own. It has answered more than three thousand questions since launch.
Open case file →Multi-agent on MicrosoftAI Value Register
An operating model for putting AI to work inside a business. Four connected agents take an idea from anyone in the organisation, decide how it should actually be built, screen it for risk, and attach a value case with a success measure the business can be held to.
Open case file →Intelligent automationClaims triage
An end to end automation that reads an inbound claim, extracts the fields that matter, checks them against the record and routes it. Built and tested against a thirty case scenario set before it went anywhere near real data.
Open case file →Agentic AI and implementation
Multi agent systems with connected agents, tool use and orchestration, including Model Context Protocol. Retrieval augmented generation, hybrid vector and keyword search, embedding pipelines, query classification and intent routing. Multi provider orchestration across Anthropic, OpenAI and Google with automatic fallback and cost tiering.
Microsoft platform
Copilot Studio multi agent solutions with connected agents. Power Automate and Power Apps running live business workflows in a global FMCG environment. Power BI, DAX and Power Query. Dataverse, SharePoint and Azure. Sitting the Microsoft AI Transformation Leader certification in September.
Value cases and benefits realisation
Building the case for an AI initiative and, more importantly, proving afterwards that the benefit landed. Effort and benefit estimated from real volumes, a named success measure and a baseline captured before anyone builds anything, and a register that shows claimed value against realised value rather than reporting green on day one.
Evaluation and assurance
Evaluation harnesses that score retrieval, comprehension and citation accuracy, with a model judging what is allowed into a knowledge base at all. Scenario test sets written before the build, so the score means something. Measured recall on KiwiStart moved from 63 to 83 percent, and I can name the changes that did it.
Responsible AI
Personal information redacted before any model call. Prompt injection and scope guards. Deterministic checks that refuse high stakes questions instead of guessing. A human stays in the loop wherever a decision affects a person. Knowing where AI genuinely helps and where it should not be used at all is the part that protects a programme.
Enablement and executive engagement
Workshops and forums with operators and executives in the same room. Training non technical people to independence and building internal champion networks. Sitting on a month end committee with the Head of Sales, customer experience and finance, which is where a technical position has to survive contact with people who own a budget.
I measure what I build, and I know what not to automate
Most AI work is assessed on impressions. I write evaluation harnesses instead, so I can tell you what changed and why. On KiwiStart that took measured retrieval recall from 63 to 83 percent, and I can name the changes that did it.
The other half is judgement. Knowing what not to automate protects a programme more than any model does. Where a decision affects a person, a person stays in the loop. Where an output cannot be checked against a source, it does not run unattended. Those rules are built into the systems, not written on a slide, and they are the reason the teams using them keep using them.
Three things a business usually needs, in this order
01 Find the work worth doing
Most organisations have more AI ideas than capacity, and no consistent way to choose between them. I run the discovery: sit with the people doing the work, map what actually happens, and put every candidate through the same triage. Some of them need a model. Several of them need a spreadsheet fixed. Saying so early is where the money is saved.
02 Build it, and prove it works
Design, build, integrate and ship, then measure it against a test set written before the build started. Guardrails go in at the same time as the capability, not after the first incident. You get a working system and a number you can defend to whoever asks.
03 Make it stick without me
Training, champion networks, written playbooks and standard operating procedures, and a benefits register that reports claimed value against realised value. An engagement that only works while I am still in the room has not been delivered.
Sectors
FMCG and consumer goods, supply chain and logistics, property and asset management, professional services and the public sector. Auckland based, working across New Zealand and Australia, and comfortable travelling across APAC.
Certification
AI Fluency: Foundations & Framework
Anthropic Academy
Certification
Microsoft Power BI Data Analyst
Professional Certificate
In progress
Microsoft Certified: AI Transformation Leader
Exam AB-731, sitting September 2026
Education
BSc Computer Science, Mathematics and Statistics
Osmania University
Changing how businesses run has been the job the whole time. At Danone Nutricia I am automating vendor invoice processing across Coupa and Basware, rebuilding manual processes as Power Apps and Power Automate workflows, and leading a New Zealand market launch. Before that, Tourism Holdings and Dell Technologies, across New Zealand, Australia, EMEA and India. The tools changed. The work did not.
That operational grounding is what makes the AI work land. At Tourism Holdings I was the person everyone queued behind for reports, so I trained the branch and operations teams nationwide to run their own instead. The win was when people stopped needing me. Adoption rarely fails on the technology. It stalls when somebody who has never done the job decides what should be automated, and nine years of doing the job is how you avoid that.