Build log

Experiments, failures, discoveries and progress from the lab.

Working notes from real prototypes. The focus is on what was learned — not a polished case study.

2026-10-28

The Action Code Nobody Was Checking

The controller honoured one command action and silently ignored the rest — and the interface sent the ignored ones anyway, because nothing rejected them.

2026-10-28

Teaching Gears to Belong to Groups

Drag a light onto a group in a browser and have it actually stick — and then read the gear back to prove it did.

2026-10-25

Running the Same Stack Three Different Ways

An engineer at a desk, a salesperson in front of a customer, and a test rig that isn't real — three surfaces over one lighting system, and the discipline that keeps them from breaking each other.

2026-10-23

Deploying an OPC UA Gateway to the Cloud

Moving a collector and historian off the office network and onto a cloud box, over a private link, to a PLC that doesn't know anyone is listening.

2026-10-10

The Hardest Part of Industrial Software Is Being Lie-able

Every operator UI is tempted to smooth over a bad reading. Resisting that temptation is most of the work.

2026-10-09

A Demo Three People Can Use Without Training

The engineering control app and the sales demo are two different products over one rig — and they can't share the same access rules.

2026-10-07

Rebuilding the Site Around the Work

The site had grown into one long dashboard with everything bolted on. Splitting it into projects, lab, build log and live demos — and what the structure taught me about presenting unreleased work.

2026-06-16

One Control Surface, Many Cranes

Turning a single-rig control page into something that drives a fleet — and the identity bug that only appears once there's more than one.

2026-05-20

Getting a Web Interface to Talk to Industrial Hardware

Notes from bridging a browser UI to a real industrial controller — what worked, what didn't, and what I'd do differently.

2026-04-15

Building Motion Into a Small AI Robot

What I learned adding expressive movement to a desktop robot — from servo calibration to motion libraries to making the robot feel alive instead of mechanical.

2026-03-10

Testing Radar for Privacy-First Presence Detection

Early experiments using mmWave radar to sense presence and activity in a room — without cameras or microphones.

2026-02-20

Lessons From Running Local LLMs on Consumer Hardware

A practical note on what works, what doesn't, and what's actually useful when you run language models locally on Raspberry Pi, Jetson and modest GPUs.