AI systems · operational automation · technical delivery

I build AI systems, and I know the operation they have to survive.

I spent eight years working in consultative sales and admissions while an interest in data took me into machine learning and software. Those two parts of my background now meet in systems I have built for a government-facing research firm, an admissions operation, and a repeatable route from prospect research to deployed AI demos.

I work where a client's messy process has to become data, code and operating practice. I can lead that conversation and build the system that follows.

84.3%: theme recall against independent analyst coding; most disagreement was additional themes in a task with no single gold standard
24h → ~10h: approximate enquiry reply turnaround across a four-country admissions operation
6 days → 1 day: first bespoke prospect demo to the second deployment from the reusable pipeline
8 years: working in consultative sales, admissions and the operations behind them

01 / Selected work

Open case files

Read the outcome here. Open the file for the architecture and full account.

01 / Regulated AIPaid engagement · pilot/testing

Stoa: sovereign consultation analysis

A New Zealand research firm needed an AI analysis pipeline designed around government security requirements without making each consultation too expensive to run.

~1 hour: AI mapping stage per 1,000 submissions
Human review: before final use
NZ$20k-30k: incumbent annual licence spend

Evidence preview · review boundary

Raw data
Up to nine steps
Analyst reviewHuman decision
Audit trail

Open the case file →

02 / OperationsAdmissions · four regions

Enquiry to enrolment automation

I connected Gmail, HubSpot, call transcripts and enrolment paperwork so information followed each applicant through to Student Services.

24h to ~10h: reply time
Same day: handover

Open the case file →

03 / DeliveryReusable pipeline

Website to demo system

A repeatable route from prospect research to an evaluated and deployed AI demonstration.

1 day: second deployment
Eval gate: before release

Open the case file →

02 / Working range

I can own the whole route

See how I have done each step in practice →

Find the problemClient interviews, workflow observation, value and risk.Sales and admissions
Map the workData, state, decision rules and failure paths.Process design
Build itPython services, model workflows, retrieval and integrations.Engineering
Test itEvals, human review, traceability and cost limits.Quality
Put it to workDeployment, training and iteration with the people using it.Delivery

03 / About

I came into engineering from the other side

Before I built software, I wrote sales reports, coached advisers, ran events and handled the admin nobody measures. That experience helps me tell the difference between work that needs judgment and work people repeat because the process grew that way.

OperatorSales, admissions and regional management
Data practitionerData science, machine learning and applied research
Software builderAI systems, evaluation and deployment