A.I.-assisted software development.
Architect-led. Not vibe-coded.

I build bespoke web applications, integrations and working prototypes for organisations that need real software fast — without the team, the budget or the timeline of a traditional dev shop.

This isn't vibe coding.

Vibe coding — a term that's gained traction in 2024–2025 — usually means asking an A.I. to write you software, accepting whatever it produces, and shipping it without really understanding what you've shipped. When it works, it feels like magic. When it doesn't (and at scale, in production, or under load, it often doesn't), the failure modes are brutal: subtle data corruption, security holes nobody saw, code so A.I.-generated that no human on the team can debug it.

I do something different.

I bring twenty-plus years of writing software, leading architectures, and shipping production systems — including more than a hundred document-capture and automation solutions for NHS Trusts, banks, insurers and councils. A.I. is a tool I use the way a senior developer uses a fast junior team: to accelerate the typing, the boilerplate, the first-draft tests and the obvious refactors. Every architectural decision, every security boundary, every integration contract is mine. Every line of code that reaches production has been read and understood by a human who knows how to debug it later.

What I build

Bespoke web applications

Internal tools, admin portals, customer-facing apps. The kind of build that used to need a four-person team and three months. I can usually scope one in a few days and ship the first usable version in two to four weeks.

Integrations & automation

The unsexy plumbing that connects your systems: API integrations, scheduled jobs, ETL pipelines, glue code between SaaS products. Frequently the highest-ROI work in any organisation, and the work nobody else wants to take on.

Prototypes & proofs of concept

Working software in days, not months. For when you need to show stakeholders something real, validate an idea before you commit serious budget, or de-risk a build-vs-buy decision with actual code in front of you instead of a slide deck.

Architect first. Prototype fast. Iterate against feedback. A.I. throughout — but supervised.

The mistake people make with A.I.-assisted development is treating the A.I. as the architect. It isn't, and shouldn't be. A.I. is excellent at the middle eighty per cent of building software — implementation, refactoring, generating tests, writing the obvious code, drafting documentation. It's bad at the hard twenty per cent: system design, integration boundaries, security, what to actually build, what failure modes matter, what's worth doing well and what isn't.

So the workflow looks like this:

It's faster than traditional development. It isn't faster because the A.I. is doing the thinking. It's faster because I'm not writing the boilerplate — and because you're seeing real software in days, not waiting for a big-bang reveal in months.

Selected projects

Three projects spanning the years before A.I.-assisted development was practical and the months since — the long-form delivery experience that the A.I.-assisted practice now amplifies.

Scanning bureau booking & billing system

For
A large UK scanning bureau (anonymised at client request).
The problem
New ownership prevented continued use of the existing internal system, but the day-to-day post-sales operation could not pause — boxes still arrived, jobs still needed scheduling, customers still expected delivered images and accurate invoices.
What I built
A bespoke web application covering the parts of the workflow that scanning software doesn't — booking, intake, scheduling, customer-side visibility, image/data delivery and billing. Integrates with both Tungsten (Kofax) and ABBYY for the scanning legs.
Where A.I. helped
Bridged logic gaps in the workflow engine and accelerated the front-end build. Core architecture predates the A.I.-assisted era.
Timeline
6 weeks from scope to first usable release.
Stack
ASP.NET, SQL Server, on the client's own infrastructure.
Outcome
Replaced the legacy system without missing a single billing cycle.

Intelligent Data Capture plug-in

For
Multiple UK accounts-payable departments running low-end scanning software.
The problem
Inbound invoice indexing was being done by hand because Kodak Capture Pro lacks the fuzzy vendor and header matching that high-end products like Tungsten (Kofax) Transformation Modules and ABBYY FlexiCapture provide. Upgrading the platform was cost-prohibitive for these clients.
What I built
A post-OCR processor that runs against scanned-and-OCRed output, matches vendors and header data using fuzzy algorithms, and feeds the cleaned index back into the existing scanning workflow. Drops in alongside Kodak Capture Pro without replacing it.
Where A.I. helped
Improved the user-facing configuration UI and the integration glue between the engine and client systems. The matching engine itself is pure highly-optimised C++, written before the A.I.-assisted era.
Timeline
Approximately one month per client deployment.
Outcome
Around 85% of inbound invoices indexed automatically — at a small fraction of the cost of moving to a high-end capture platform.

Mobile application

For
Personal side-business / portfolio.
The concept
An as-yet-unreleased logic puzzle — combines logical deduction with reflex-led play.
What I'm building
Native Android in Java with custom OpenGL ES shaders. A.I.-assisted throughout, from architecture and shader work to UI polish.
Where A.I. helped
Compressed several years of solo evening-and-weekend work into months — effectively the throughput of a ten-person team without the team. This is the project that most directly demonstrates what the A.I.-assisted practice can do for a one-person shop.
Timeline
Six months in active development.
Status
Private beta with two testers. Public launch expected Autumn 2026.

Twenty years of shipping software, supercharged.

A.I. didn't make me a developer. I've been one since 1981 — programming as a hobby through school, professionally since the early 2000s, then twenty years leading complex document capture and automation projects for some of the UK's largest organisations.

What A.I. changed is throughput. The same architectural instincts, the same code-review discipline, the same "will this still work in three years?" habits — but applied across more code, more projects, and more delivered software per week.

If you want vibe coding, there are cheaper places to get it. If you want software that ships, runs reliably, and someone can still debug in two years' time, this is what that looks like.

Got something you'd like built?

Send me a few lines about what you have in mind — even a fuzzy idea is fine. I'll reply with an honest read of what's involved, what isn't, and whether I'm the right person for it.