Welcome

First post, so it may as well say what the rest are for.

I spend my days building AI systems — document extraction, retrieval, agents — for organisations where confidentiality and traceability matter as much as the output. Almost everything I learn doing that never leaves a private repository or a video call. This is the place meant to change that, a little.

What I plan to write about

Three things, roughly.

First, the gap between a demo and a system that holds up. A demo has to work once, in front of someone. A system has to work every day, on badly scanned documents, at three in the morning, with nobody there to catch it. That distance is longer than it looks, and it is where the interesting work happens.

Then the boring decisions — the ones that never show up in a screenshot. Where the data lives, what gets logged, what a model is allowed to trigger on its own, and the exact point where a human has to say yes. Those constraints don't slow the design down; most of the time they are what makes it sharp.

And occasionally something smaller: an implementation detail that cost me a day, or a belief I held and no longer do.

How

Short, and only when there is genuinely something to say. No publishing schedule, no twelve-part series. If a post can be one paragraph, it will be one paragraph.

I write from what I have built, never from what I have read elsewhere. Client specifics stay out — most of it is under NDA — but the lessons generalise, and those are the part worth keeping.

Why now

Because I am optimistic enough about this technology to want it talked about properly. Done well, it makes people more capable; done badly, it costs them their trust in tools that could have helped. Writing down what I actually see seems like an honest way to push in the right direction.

Thanks for stopping by. More soon.