Home/Blog
Notes from the lab
Working write-ups from the people doing the work — what we tried, what it cost, and what we would do differently.
Latest
Why we scope the deployment target before the model
Most machine-learning projects that fail in the field were scoped against a GPU that will never leave the office. A note on starting from the power and latency budget instead.
Read the write-upMore Posts
Everything else we have written up
What a first-spin board actually costs you
Every respin is four to six weeks. A look at the review steps that catch the errors worth catching, and the ones that only make the schedule longer.
Read
Documentation is a deliverable, not an afterthought
How we structure decision records, test evidence and hand-over packs so a client team can own a system without us in the room.
Read
Working with imagery when the sky is never clear
Field notes on building change-detection pipelines for regions where optical coverage disappears for a third of the year.
Read
Threat modelling without stopping the programme
Security reviews get skipped when they arrive as a gate. What it looks like to run them as part of architecture instead.
Read
Designing for the network you actually have
Edge systems get specified against a lab network and deployed on a link that drops every afternoon. What we build in from day one so the system degrades gracefully.
Read
New write-ups land every few weeks.
We do not run a newsletter yet. Until we do, the quickest way to hear about new work is to ask.