Rebuilding myself as an AI systems engineer.
I'm Rich Woollcott — a software engineer of 25+ years (contracting since 2001), based in the South West, UK. I've spent my career shipping mobile, backend and desktop software for companies like SSE, De La Rue/Crane, ClearAccept and DHL/Evri.
In 2025 I started rebuilding myself as an AI systems engineer — and decided to do it in public. Not the "top 10 AI tools" kind of AI content: the unglamorous layers that actually make AI systems work, and that stay scarce when everyone has access to the same models:
Agent harnesses
Loops, tools, memory, validation, quality gates and review; making agents reliable rather than impressive.
Datasets and evaluation
Generating training data, judging quality, and knowing when fine-tuning is worth it.
The serving layer
Cost, latency, throughput and local inference; I run fine-tuned models on an Nvidia DGX Spark on my desk.
Agent ecosystems
Planner, architect, implementer, reviewer and QA agents co-operating over a message bus — a software factory, which I treat as my lab bench.
Physical AI
Agents meeting robotics, starting with a Reachy Mini that tutors my daughters for their GCSEs.
Good places to start
- The YouTube channel — the Study Tutor robot demos and my DDD South West talk on software factories.
- From Spec-Driven Development to Feature Plan Development — the thinking behind GuardKit.
- guardkit.ai — docs for the open-source toolkit, and the wider fleet on GitHub.
- My fine-tuned models on Hugging Face.
Work with me
Through Appmilla I take on AI engineering and AI-augmented delivery work — recent examples include an AI-monitored open-banking platform for a fintech, an AI legal-research proof of concept, and rolling out AI-assisted development (governance, steering files, training) across a national logistics company's engineering teams.
Book a call
Thirty minutes to talk through where you are.