Notes from the work
What we have learned on projects, written for the people who run operations. Practical guides you can act on, field notes from the engineering, and sector pieces.
Practical guides
- Automation, assistant or agent? A short decision guideThe three words get used interchangeably in sales meetings. They are different systems at different prices. Five questions to tell which one your problem actually needs.14 Oct 2026 · 5 min
- Is your plant data ready for predictive maintenance? A 10-question checkBefore anyone sells you sensors, answer these ten questions about the data you already keep. Most plants can answer eight of them with a yes.10 Oct 2026 · 6 min
- Fine-tune, prompt, or train from scratch: what we actually chooseThree ways to get a model for your problem, what each costs, and the plain rule we use to pick between them on a real engagement.28 Sept 2026 · 6 min
- What five years of maintenance registers can tell you (and what they can't)The register is the most underused document in the plant. A guide to reading it as a dataset, and the limits of doing so.21 Jul 2026 · 6 min
Field notes
- An agent is only as good as its harnessWhy we build the test set before the agent, what goes into it, and how it stops a cheaper model from quietly breaking a flow that worked.20 Oct 2026 · 7 min
- Digitising ten years of registers: what we learned about handwritingNotes on turning a paper archive into a dataset. Where models read handwriting well, where they do not, and the validation step that matters more than the model.5 Oct 2026 · 6 min
- Why the first six weeks of every AI project is data cleaningA plain account of where the time actually goes in an operations ML project, so that nobody is surprised in week five.22 Sept 2026 · 5 min
- Running ML on-premise in an Indian factory: what actually brokePower cuts, a server under a desk, and a plant IT engineer with twelve other jobs. Notes on keeping a model alive where the cloud is not an option.2 Sept 2026 · 7 min
Sector pieces
- The data a university already has to spot at-risk studentsAttendance, assessment and LMS activity predict end-of-term trouble weeks ahead. Most institutions already collect all three.18 Aug 2026 · 5 min
- When a rule beats a model: three cases where we did not use MLThree engagements where the best engineering answer was a control limit, a report, or a form. Why we recommend them, and how we tell the difference.4 Aug 2026 · 6 min
Not sure if your data is ready?
Take the free 20-minute readiness check. You get a short scorecard and a plain answer about what, if anything, is worth doing next.