I build AI that understands how money moves.
I'm Head of Product at Paysecure, a payments orchestration company. Before that I led payments product at Entain, across more than 20 PSPs and several regulated markets.
Outside work I build AI products end to end, test them on real people, and write down what I learn. This page is the record.
- Now
- Building Nura, in private beta, and EvalDesk, in the open.
- Open to
- A few advisory conversations on AI in payments. Work with me
What I'm building
Each one started with a problem I had myself. Status is honest.
-
PRIVATE BETA
Nura
A WhatsApp-first family AI assistant for busy parents. It keeps a shared memory of school, classes and events, and sends both parents the same morning update.
What I learned: memory is the product. Concise, reliable messages matter more than clever ones, and my first user (my wife) told me so quickly.
WhatsApp · AWS serverless · Supabase · OpenRouter · golden-set evals
-
LIVE
StayOnPoint
Stops you rambling in interviews. Practise an answer out loud and live cues keep you on point: headline first, key points covered, then stop. You get a summary after each answer.
What I learned: a small, fast classifier beats a large model when the job is a quick yes or no in real time.
Deepgram · Jev classifier · GLM via OpenRouter · real-time audio
-
IN PROGRESS
EvalDesk
An open-source review desk for subject-matter experts. Read your AI product's conversations, find its failure patterns, and build checks you can trust. It runs in your browser.
Why: the people who know whether an AI answer is right are often domain experts, not engineers. Evals should not need code.
Open source (MIT) · runs in the browser · payments demo
-
LIVE
The AI Hotel
Short YouTube explainers that teach AI concepts through one recurring hotel. The model is the concierge; evals are mystery guests; MCP is a plug socket. First series on my channel, It's Like This.
Why: I learn best through analogies, and most explainers skip where the analogy breaks. Every episode includes that part.
React animation kit · ElevenLabs narration · 2–3 minute episodes
-
IN PROGRESS
Payments context kit
An open-source, vendor-neutral context layer that helps any AI reason about card declines. Version 0 is a decline-event schema, a reusable question pack and a golden test set.
Why it matters: models are interchangeable. The payments context you give them is not.
Open source · model-agnostic · declines first -
LIVE
TabSearch
A Chrome extension for people with 50+ tabs open. Search any tab as you type, see how old your tabs are, and close duplicates in one click. On the Chrome Web Store since January 2026.
Privacy first: fully local, with no tracking, no account and no network requests.
Chrome extension · 100% local storage
Also built: an AI payment compliance dashboard and a privacy and compliance new-tab browser extension.
What I'm writing
Short write-ups from the builds above. First pieces are on the way.
- Rules, a classifier or an LLM? Picking the right brain for payment routingDRAFTING
- Why every payments AI needs a context layer for declinesDRAFTING
- Evals for people who don't write code: lessons from a family agentDRAFTING
- What a WhatsApp agent taught me about memoryDRAFTING
Where I've worked
I've worked across the whole ecommerce journey: checkout and payments at ASOS and Harvey Nichols, fraud integrations at Signifyd, the full payment stack at Entain, and orchestration at Paysecure. I know 3DS, PSP routing, decline handling and regulators like the UKGC from the inside.
I studied computer science, have an MBA, and I'm comfortable in SQL and APIs. These days I build with Claude Code and Codex most evenings.
Deploying AI in payments?
I help payment teams, vendors and merchants, take AI from demo to something they can trust: the right use case, the payments context, and evals that tell you when it's working.
Open to a few advisory conversations. Tell me what you're working on.