Applications · Automation · AI governance
I've spent twenty years in education, as a senior leader, a trainer of governors and the chair of a governing board, and I build the technology governance work needs.
I founded SGAF (opens in a new tab), the governance platform for UK schools and academy trusts, and I design, build and run production web and mobile apps with a team of AI agents built specifically to be held to the same governance standards I teach. My work sits where the two meet: software shaped by someone who has sat in the meeting, and AI put to work on the problems schools have, not on demos.
On this page you'll see how I do it, and the process behind it.
Four areas I lead. Each links to work you can open, and the evidence page has the rest.
Know what the estate is for, cut what duplicates, and build only what earns its place.
See it in
Prioritised, governed automation connected to the official data and product events it depends on.
See it in
Every change gated before it ships, the automation monitored, and outages that reach a person.
See it in
Clear decision rights, checks that can fail, and data protection that schools can rely on.
See it in
Eight featured here. The full index adds case studies and notes, from a health coaching loop to a Raspberry Pi wired to an agent, each with the problem, what I built and the result.
Governance assurance for UK schools and academy trusts. Audits, risks, visit plans and the annual statement on one connected record, so the board does the governance once and the evidence is created as it goes. More than 500 schools have used its free tools.
Every state-funded school in England, in context: inspections, attendance, results, pupils and teachers beside the local authority and England, rebuilt and checked every week, on a map that works from the whole country down to a street. Nothing on it ranks a school.
A bespoke HTML presentation for each client meeting, served from the SGAF portal itself, so a demonstration moves between the argument and the live product without changing windows.
Around forty-five single-purpose agents behind one front door, running radar, outreach, meeting prep, support, product, content, data quality and finance. None of them send, spend or delete.
My own scheduler, built where the product facts and the content catalogue already live, so posts know what the business knows. Fingerprinted approvals, a brand gate on every draft, and a dry-run week before anything goes live.
Courtside netball scoring, built for my wife, who umpires netball. Prepare the match, score live, track the centre pass, share a clean result at the final whistle. Native, because the job happens standing up in bad light.
A free prompt library for school staff, early years to sixth form and the board. Fill in your school once, copy the prompt, run it on a machine you control, and a person still owns the draft. The flag that marks a record as never for a public chat started here.
A daily puzzle. Four gems, a reading that counts set and loose, a new seal at midnight. No account and no server. It stays on your device, which is exactly what I wanted to learn how to ship.
Markdown posts, a static build, social cards generated per post, an RSS feed my agents can read, and a copy button that turns any post into an article for X or LinkedIn.
In the order to read them. Each one stands alone and republishes as an article on X or LinkedIn with one copy button.
Notes from a demonstration I gave to a successful edtech company: how a fully autonomous operation actually runs, why a new company can work this way from day one, why an established one has to manage it as change under board pressure, and why the answer is a place to play.
Edtech sells headcount. I run a governance product for UK schools on my own, with a team of AI agents that prepare everything and decide nothing. Here is how the pipeline works end to end.
A restructure I had put off for months: how the team got to forty-five, what the overlap was costing, the weekend I spent deciding who stayed, and the part I left out until the end.
An agent will approve its own output every time. So I never let it. Every agent that produces something has another whose job is to find what is wrong with it, and the last gate is always a person.
A scheduling tool knows when to post. It does not know that a product name was retired last week, that a counting claim is banned, or what schools are doing in March. Mine does, because it sits next to everything the business knows.
A daily puzzle, a courtside scoring app, and a free prompt library for school staff. None of them were built to make money. All of them changed how I build the thing that does.
Help articles that match the live product, explainer films inside them, a teaching skill that checks every claim against Ofsted and the DfE, and a weekly freshness sweep. Current is a property you test, not a promise you make.
Most of what a school data dashboard shows is published free by the DfE and Ofsted. I built School Intel to put every state-funded school in England in context, rebuilt and checked every week, for almost nothing. Before you renew a dashboard, ask what you are paying for.
Send a message, or pick another route. Every reply comes from me.
SGAF for your school or trust
Writing and updates
