Software Engineer · Founder, Northwind Labs
I build production software and mathematical models (from a tamper-evident evidence app for renters to local LLM deployments and market-analysis engines) for people and businesses that need results they can verify.
Computer Science & Mathematics Northwind Labs Ltd London, UK
Joint-honours in Computer Science and Mathematics. My work sits where provable maths meets shippable software.
I design and ship production systems end to end: architecture, security, data protection, and deployment. I care about systems that are fast, correct, and measurable, and about explaining what they do in plain terms to the people paying for them.
Three problems I solve, described without jargon.
On-prem · RTX 5080
Private AI that never leaves your building. I stand up local language models on high-end GPU hardware so you can analyse sensitive data (customer records, pricing, contracts) without sending any of it to a third-party cloud.
Dealership analytics
Find the profit already sitting on your forecourt. I score your stock against live market data to flag undervalued vehicles and high-margin specifications before they’re mispriced.
Operations research
Scheduling, routing, and resource allocation solved as maths, not guesswork. Custom models that cut cost and cut waste.
Three systems, built end to end.
Live in production
A mobile-first web app that helps private renters in England build a defensible, time-stamped evidence trail (condition photos and repair logs) and export it as a court- and adjudicator-ready PDF.
The Renters’ Rights Act 2025 passed, but its new Ombudsman isn’t expected until roughly 2028, leaving a multi-year gap where tenants have no simple way to document issues and escalate correctly. Free core tool, plus a £12 one-off “Dispute Pack.” Built solo, end to end: product, architecture, compliance design, deployment, and ad operations. Operated as Northwind Labs Ltd, an ICO-registered data controller.
Stack
Engineering highlights
getUserMedia and
fingerprinted client-side with a SHA-256 hash. The capture method and hash are
printed into the exported PDF, so a third party can verify a file hasn’t
been altered since capture.ffmpeg -ss./r/<slug> redirect routes billboard QR scans through UTM
tagging into GA4, making physical out-of-home advertising measurable
(normally a black box).By the numbers
Built · MetaTrader 5
An algorithmic trading system for XAUUSD, written as a MetaTrader 5 Expert Advisor. It buys pullbacks inside an established trend, sizes every position from its own stop distance, and stands aside once a move has gone parabolic.
Most trend systems fail on the same two things: they size positions by habit rather than by risk, and they enter hardest exactly when a move is about to exhaust. This one derives its lot size backwards from the ATR stop, so every trade risks the same fraction of equity whatever the volatility, and it treats a high ADX reading as a reason not to trade rather than a confirmation.
Shipped · v4.1
A production Python engine that watches used-car markets for price dislocations, verifies each candidate against DVLA records and multi-round vision AI, and surfaces only the vehicles worth acting on.
The hardware behind the on-premises pitch, stated plainly.
If you have a problem that looks like maths, or data you can’t send to the cloud, I’d like to hear about it.
Start a consultation