Things I've taken from idea to production.
Each of these was a real product or programme I led, for governments, banks, and global brands. Same pattern every time: name the problem honestly, ship something people actually use, and put a number on the result. Further down: 17 more client programmes published by Hauraki, the New York AI consultancy whose delivery organisation I built and scaled.
The problem
City planners and developers sit on mountains of spatial, social, and economic data and still can't get a straight answer out of it. A single question (where should this facility go, and who does it affect?) meant weeks of analyst time, and the answer came back without anything to back it up.
What I did
I led product and delivery on an AI platform that fuses spatial, social, and economic signals into plain-language answers, and cites the source behind every claim, so a planner can defend the decision. I built and ran the delivery team, set the QA bar, and kept the thing honest enough to put in front of government.
The result
The problem
Dietary advice is generic, but diets aren't. Matching real supermarket products to one person's allergies, conditions, and goals means scoring hundreds of thousands of SKUs against messy nutrition data, and no one had cracked it at consumer scale.
What I did
I built the data-science engine behind Bitewell that matches food products to individual dietary requirements. That engine became the core technology of what is now FoodHealth, and carried the company's pivot from a food app into a food-health enterprise.
The result
The problem
Enterprise recruiting still runs on keyword matching. Strong candidates get filtered out because their last job title doesn't match the req, while recruiters wade through thousands of lookalike CVs to find the few who can actually do the work.
What I did
As lead Product Manager I delivered skills-based matching and candidate-screening workflows built on Eightfold's talent-data engine, so the system ranks people by what they can do and what they could learn, not by the words on their CV.
The result
The problem
Brands almost always react to a trend after it's already peaked. By the time something shows up in a standard dashboard, the window to do anything useful with it has closed, and the budget's been spent chasing last month's story.
What I did
I shipped a generative-AI platform that surfaces emerging consumer and cultural trends 3–4 weeks ahead of the market, took it from concept to a paying product, and scaled it across three very different industries without rebuilding it for each.
The result
The problem
Regional brands fly blind on how they're perceived, especially in Arabic, where most Western tools flatten dialect and sarcasm into nonsense. A crisis is usually only visible once it's already trending, which is the worst possible time to find out.
What I did
I shipped a real-time platform that tracks sentiment, share of voice, and competitive position across 100+ languages with native Arabic understanding, and pushes crisis alerts the moment something turns, not the morning after.
The result
The problem
Most marketing still slices people by age, income, and postcode, labels that say almost nothing about what someone will actually buy. Teams build campaigns on segments that are, frankly, made up, then wonder why the targeting misses.
What I did
I set the product strategy and built the initial MVP for an AI tool that defines consumer segments from real behaviour rather than demographics, so marketing and product teams target people by what they do, and can act on segments that actually predict something.
The result
The problem
Restaurants throw away a brutal share of what they buy, and over- or under-staff almost every shift, because POS, inventory, labour, and recipes all live in tools that don't talk to each other. Nobody can see what tomorrow actually needs.
What I did
I shipped an AI platform spanning POS, inventory, labour, and recipe engineering, with predictive models that forecast demand and flag waste before it lands in the bin, then scaled it past the pilot into real estate it had to hold up in.
The result
The problem
Local delivery is three groups (couriers, customers, and restaurants) all on different systems and all annoyed at each other. Wrong orders, late drops, no visibility, and a reputation that erodes one bad delivery at a time.
What I did
I shipped an AI-enabled platform that connects all three sides of the delivery marketplace, with matching and routing that put the right courier on the right order, and grew it from a handful of restaurants into a real network.
The result
The problem
Staffing consultants across hundreds of live engagements was done by hand: slow, error-prone, no real-time view of who was actually free. People got double-booked, projects waited, and nobody could see the whole board at once.
What I did
I directed the programme that replaced manual assignment with an automated allocation engine, giving leadership a live, single-pane view of capacity across teams, and built it to fit how the firm already worked rather than forcing a new process on everyone.
The result
The problem
Global product data lived in a dozen inconsistent taxonomies, so SKU-level forecasting was unreliable and categorising new products was painfully slow. Every team trusted its own version of the truth, which meant there wasn't one.
What I did
I oversaw the global data programme: standardising the taxonomy and rebuilding the forecasting on top of it, so predictions were trustworthy and new products slotted into the right place automatically instead of by hand.
The result
And a few more I've shipped.
Same story, less ink. Products and programmes I led across fintech, recruiting, and food-health.
AI crypto investment platform
Ran the rollout of an AI-driven investment platform: predictive analytics, compliance automation, and secure trading-API integration.
Conversational AI wallet
Directed a wallet that pairs payments with AI chat for peer-to-peer transfers, inside a secure compliance framework.
US digital-banking transformation
Led roadmap execution embedding AI-powered KYC automation, AML monitoring, and cross-border transactions into a digital bank.
17 more engagements, published by the consultancy I helped build.
At Hauraki, a New York AI consultancy, I built and scaled the delivery organisation of data scientists, engineers, and PMs behind the firm's client work. Some of these I led directly; all of them shipped through that team. The write-ups below are Hauraki's own published case studies, so you don't have to take my word for the numbers.
Nike: Multilingual Search & Taxonomy LLMs
Fine-tuned LLMs for product search and taxonomy across 26 languages and global markets.
EY TaxTech: K-1/K-3 Document AI
Document AI and smart queueing for tax-form extraction at a Big Four firm.
PwC: Private-Equity Alpha Finder
A RAG swarm that cut deal research from 5-7 days to 2 hours.
BlueCross BlueShield: Claims Automation
GenAI suite that automated 35% of claims processing for a leading US health insurer.
Saudi Customs: Post-Clearance Audit AI
Policy-aware RAG accelerating customs audits across the Kingdom.
NZ Customs: Cargo Risk Detection
Risk ML for container screening, awarded a WCO Certificate of Merit.
Trinidad & Tobago: Revenue-Evasion Predictor
GraphRAG flagging under-declared customs duties.
ACC Insurance: Provider-Fraud Scoring
Behavioural ML micro-services replacing legacy fraud systems.
Nivelo: Real-Time ACH Fraud Detection
Low-latency fraud engine that became a licensing revenue stream.
Healthdex: Private Health-Data Marketplace
Homomorphic encryption for privacy-preserving health data exchange.
BioSynapse: Drug-Discovery Knowledge Graph
Agentic knowledge graph accelerating drug-target discovery.
UBF: Donor-Impact Analytics
GraphRAG for impact demonstration and grant optimisation.
Auckland Blues: Injury Prediction
Multimodal edge AI predicting soft-tissue injuries.
Centrality: Crypto-Trading AI
Multi-signal ML for ICO screening and trading.
MASI: Low-Resource-Language Translation
Fine-tuned LLM translation bringing bilingual education to rural students.
FactoryFlow: Voice-Assisted Maintenance
Predictive maintenance with voice AI for technicians.
AmplifyInfluence: Creator Campaigns at Scale
Multimodal generation for hyper-personalised influencer marketing.
Want your project on this page next?
Tell me what you're trying to do. A 30-minute call, no pitch. I'll tell you straight whether it's worth doing, and how I'd ship it.