AI Engineer · Automation · Agentic Systems
For teams automating complex, manual processes — from the first diagram to production, with human review where the decision matters
Trusted by
Simple KYC 
Sapira.ai 
Hastings Direct 
NTT Data 
Gonvarri 
King's College London 
University of Kent Delivered solutions for
American Express 
Westpac 
Telefonica 
Tirea AI systems that solve real problems
End-to-end pipelines that run without human intervention. Research, coding, deployment, and review workflows orchestrated by multi-agent systems.
Interactive assistants embedded in Slack, web apps, or any workflow. Connected to real tools — databases, APIs, infrastructure — to act, not just answer.
Specialized applications for regulated industries and complex business domains. Classification, document parsing, risk assessment, and full product development.
The integration layer powering everything above. Model Context Protocol servers connecting LLMs to Bitbucket, PostgreSQL, Langfuse, Azure, Kubernetes, and more.
Three doors into the same work: turning something that works at small scale into a system you can depend on.
01 Product
Where it gets stuck
You have a product in production and an AI feature that shines in the demo and falls over on real data. Nobody can say why it answered what it answered.
How I come in
I come in to make it measurable before you open it to everyone: evaluation on every change, cost and latency limits, and the traceability to explain each answer.
02 Own systems
Where it gets stuck
You have an ERP, a catalogue, tickets or documentation the business depends on, and a layer of manual work on top that nobody has managed to remove.
How I come in
I connect agents to those systems respecting your permissions, with human control where the decision matters and a clean exit if you want to switch it off tomorrow.
03 No technical team
Where it gets stuck
You know time is being lost, but not whether yours is automation or AI, nor where to start, nor what it should cost.
How I come in
I start by telling you which of the two you need, even when the answer is the cheap one. No jargon, and no commitment until you have seen the map.
Reusable agent workflows I package as Claude skills — the layer I reach for before MCP.
Records narrated product-demo and verification videos of any web app by driving a real browser and assembling beat-narrated MP4s.
Guides new-feature development: architecture planning, implementation and PR creation.
Audits a repository against a Responsible-AI checklist, spawning parallel agents to close code-actionable gaps and tracking compliance.
Less hype. More working systems.
End-to-end pipeline: research → specs → code → deploy → verify with automated feedback loops and self-healing.
50+ microservices shipped through the pipeline, with no manual step
Slack-based AI agent for infrastructure diagnostics. Queries Kubernetes, Azure, CI/CD pipelines, and documentation autonomously.
70% of routine requests resolved without a human stepping in
Model Context Protocol server for Bitbucket API operations. Manage repositories, pull requests, and code reviews through LLMs.
58 tools covering the whole API, published on npm and PyPI
Full pricing management system for a steel industry corporation. AI assistant, email-to-table updates with approval, and PDF tariff digitization.
925+ live price tables from 7 suppliers, every change human-approved
Industry classification with risk assessment. Official registry connectivity and adverse media screening.
<1 min per entity, against the 10–30 minutes it took by hand
Live product: a solar vitamin D calculator PWA in six languages. Computes synthesis windows from solar geometry, live UV and skin type — and exposes the same engine to agents over MCP.
6 languages in production, installable PWA and open source
Times are effective working time across the whole project, not calendar time and not a single phase: most went through several functional phases, with pauses and shifts in priority along the way.
I don't sell a closed product, I sell a method. These are the five phases every project goes through, and what you walk away with from each.
One call to understand the process as it actually runs today: who does it, what it costs, and where it breaks.
An honest read on whether it's worth automating
I draw the system before writing it: which agents, which data, which integrations, and where human review belongs.
Diagram and technical decision record
Built in vertical slices: each iteration leaves something working end to end, not a loose module. A large project is several phases chained together, each one deliverable on its own.
Weekly drops you can actually try
We decide with your team what a person checks and what can run unattended. That line gets measured, not assumed.
Measurable quality criteria and evals
It ships with monitoring, and your team gets the code and the documentation to keep it running without me.
Code you own, documented and deployed
If your bottleneck is fixed by a SQL query and a cron job, I'll say so and we won't build agents. AI only goes where it earns its place.
Every system I build has an explicit human checkpoint on the decisions that matter. No black box signing off on your behalf.
Something is in production solving a real part of the problem before the first month is out. No six-month blind integration.
Three ways to start, depending on how clearly you can name the problem. These aren't fixed packages — they're entry points.
When you know there's repetitive work, but not which part to automate first
The document is yours whether or not we work together after it.
Free call · audit from £150
When you already know which process hurts and want it solved
Scope and price agreed before we start, once the diagnostic is done.
From £600, depending on scope
When your team already builds and what's missing is technical judgement
Built for teams who want to stop depending on me, not the other way round.
From £400/day
So you have a sense of scale before the call. The final number comes out of the scope we agree in the diagnostic.
What people usually ask me. If yours is not here, write to me and I will answer on the next working day.
You do, from day one. It lives in your repositories, not mine, and it gets documented as it is built rather than at the end. I work with the minimum access needed, revocable at any time.
With me. The same person who takes the first call designs the system, writes the code and answers when something breaks. No commercial layer translating between whoever sold it and whoever builds it, and no account rotating between profiles: what is agreed on the call is what gets built. So that it does not hang on my calendar either, the code and the documentation live in your repositories from the start and delivery happens in slices that already work, so there is never a stretch of months of half-finished work.
Yes, with an NDA signed before we start. I work inside your delivery under your brand, and the commercial relationship with your client stays yours: I do not touch it before, during or after.
Almost never. The diagnosis starts by reviewing what already exists: what it actually does, on what data, and what breaks when the data changes. Often what is missing is not a rebuild but everything between a PoC and something you can operate — evaluation, limits, traceability, deployment.
It happens more often than you would think: a lot of what gets called AI is automation, and the difference matters because the cost and the fragility are not the same. In the diagnosis I tell you which of the two you have, even when the answer is the cheap one. If it turns out to be automation, I build that too — it just falls outside the three ways of working above and gets quoted separately.
The discovery call is free. From there, three ways: an audit from €180 that you keep in writing even if we do not work together, a scoped project from €700 depending on scope, or ongoing support from €480/day with no lock-in. The project price is fixed after the diagnosis, once we know what we are working against.
Architecture, one or two weeks. Building, between two and eight depending on scope. Deployment and handover, one more. What does not change is that within the first month something is in production solving a real part of the problem — not six months of blind integration and a single delivery at the end.
Book a free 30-minute discovery call to discuss your project