AI & Automation Lead · Technology & Operations
Build the AI.
Run the business.
I'm an operator-engineer. I run a UK services business — and build the production AI & automation it runs on.
I'm an operator-engineer. I run a UK services business — and build the production AI & automation it runs on.
I build the production AI a business runs on — and I run the business it's built for. Most engineers optimise the system in front of them. I get to choose which system is worth building, because I'm also the one reading the P&L it feeds.
The scarce skill stopped being syntax. Anyone can generate code now; what's rare is knowing which code is worth generating — and being able to tell when it's confidently wrong. The bottleneck moved from typing to judgment, and judgment is hardest to fake in the one person who has to live with the result.
The middle is thinning. AI compresses the layer that only passes work along — the coordinators, the spec-handlers, the analysts who escalate the decision upward. What's left is people close enough to build and senior enough to own the outcome. I never sat in the middle — running the business while building the AI it depends on left no one to hand the judgment to.
A small team now ships what a department used to. AI extends what one person can actually execute, not just sketch. So the advantage goes to whoever can point it across a whole business and prove the number moved afterward. I'm not reacting to that shift — it's the thing I build.
Two sides of the same job — I build it, then I run it.
Production AI, not demos — and I prove it works.
I carry the number, not just the code.
I think in the numbers that run a business — and build the AI that moves each one. Leads → conversion → value → margin → profit.
Depth isn't ten projects named. It's one, shown completely.
The retrieval engine behind day-to-day operations was nearly blind — about 0.05 recall. It didn't need a bigger model. It needed someone to measure it and admit it was failing.
So I rebuilt retrieval as a hybrid system — vector plus keyword, with reranking — and engineered it to trustworthy, not just plausible. The model was the easy part; the measurement was the job.
Recall climbed to ~0.85, gated by an automated eval suite and a human-in-the-loop critic that refuses to auto-trust new facts. Under real audit and board scrutiny, that's not optional — it's the build.
Production systems in daily use. Figures banded for confidentiality.
An organisation-wide retrieval system that turns scattered business knowledge into grounded, cited answers — Postgres + pgvector, hybrid retrieval and reranking, an automated eval harness and a human-in-the-loop critic gating what's trusted.
Co-pilots in daily use by leadership — cashflow forecasting, an operations risk co-pilot, and marketing-attribution dashboards.
A central pricing engine as a single source of truth, powering AI-assisted instant quoting and self-service booking.
An idempotent, rate-limit-aware layer connecting CRM, field-service, finance and e-sign — replacing a fragmented manual process with automated, gate-based tracking and one live control panel.
Built the unit-economics model and board-level monthly management reporting; restructured incentives and helped take the business from owner-dependent to system-led.
What hiring me actually buys — a diagnosis, a shipped system, and a number that moved.
I read the numbers before I touch the stack. Where's the money leaking, and which single lever moves profit most? You get a diagnosis of the binding constraint — not a tool wishlist.
One production system, built at that lever and measured against a baseline — the first automation live and real hours handed back to the team.
Evals, guardrails and a repeatable cadence, so it keeps working after I stop watching it — a measurable quality number on every change, not a demo that quietly rots.
The discipline behind AI a business can rely on.
Golden-set evals, regression checks and grounding gates — reliability is measured, not assumed.
Every build starts from the business outcome and has a number it's meant to move — and I own whether it moved.
Owner-dependent → system-led: documented processes, governance and automation so nothing depends on one person.
Hands-on from model to production.
Open to AI & Automation Lead, Head of AI & Operations and Solutions Architect roles — permanent or fractional.