AI that ships, holds up, and scales.
I build AI-native products, go looking for the ways they break, and scale the systems underneath them. One engineer who has done all three in production, working directly with your team.
| Area | What I do | How long |
|---|---|---|
| AI product engineering | Build it | Two to three months |
| AI-native security | Break it, then fix it | A few weeks |
| Application scaling | Work with your team | A day or two a week |
| SaaS security | Go through it properly | A few weeks |
| AI governance | Set it up, hand it over | Depends |
Where teams get stuck
The demo worked. Now it has to run every day against real users, and nobody is sure what the agent is allowed to touch, what happens when the model is confidently wrong, or who gets paged at 3am when it is.
The product is growing and the architecture that got you here is the thing slowing you down. Queries that were fine at ten thousand rows are not fine at ten million, and the fix keeps getting deprioritised.
An enterprise customer sent a security questionnaire. Somewhere in it is a question about tenant isolation, or about what your AI features do with their data, and the honest answer is that nobody has checked.
What I work on
AI product engineering
Agentic systems, RAG pipelines, MCP servers and evals. I write the code, not a spec for someone else to build from.
AI-native security
Prompt injection, agent permission boundaries, tool-call blast radius, data leakage. The failure modes classical AppSec misses.
Application scaling
Microservices, event-driven architecture, database and latency work, so the system survives the growth you are planning for.
SaaS security
Multi-tenant isolation, IAM, zero-trust access and the security questionnaire your enterprise deal is blocked on.
AI governance
Model inventories, risk classification and EU AI Act readiness, sized so it does not stop you shipping.
Things I have built
Each of those maps to something running in production, not to a certificate:
- An agentic operations platform that investigates incidents and fixes them, behind a policy engine that stops and asks a human before anything consequential. Durable workflows, event-sourced audit trail, decisions scored by an LLM judge.
- An MCP server exposing 240+ platform APIs to external developers and AI agents, which is entirely a question about what you are willing to let an agent do to production.
- A two-region payments platform on Go microservices, with a change-data-capture pipeline behind analytics and reconciliation.
- Zero-trust access and incident response across a fintech estate, including leading the response to a wallet-custody security incident.
Selected writing
Build a self-hosted AI agent
An agent that runs shell commands, reads logs and raises pull requests, and what you take on when you give one that access.
How to become an AI engineer
RAG, vector databases, agents and MCP servers — what a backend engineer actually needs to learn, and what they can skip.
Load balancing for real traffic
How requests actually get distributed, and the failure modes that only show up once you have more than one server.
Not here to hire anyone?
A lot of people find this site through the videos rather than because they are buying engineering work. If you are an engineer trying to work out which language to learn, whether to move into AI, or why your promotion has stalled — most of that is answered free.
Who you would be working with
I am an engineering manager who still writes code. A decade across cross-border remittance, logistics, real-money gaming and government systems — places where being wrong had consequences beyond a bad sprint.
I run engineering for a payments platform day to day — team, sprint cadence, code review, release. Anything I suggest to you is something I have to live with on my own roadmap.