Loading…
Loading…
What I test before it goes anywhere near a job someone depends on — which new techniques are ready for a trade or industrial operation, and which are still a demo.
Anything new gets tried here first, on sample data and my own time, rather than on a job somebody is depending on. Most of it does not survive that, which is the reason for doing it in this order.
The three demos on this site came out of that. They run on made-up sample data, they are rate-limited, and you can open any of them right now without talking to me first.
Book a callWhat I am watching, how ready each one looks, and roughly when it might matter for a trade or industrial job. Nothing here is advice to go and buy something yet.
One model that reads text, images and audio together. It is why a photographed delivery slip and the voicemail about the same job can go through one system instead of two.
Chips built more like a brain than a processor, aimed at running models on a fraction of the power. Interesting for the day inspection runs on the machine rather than in a data centre. Not yet something to plan around.
Training a model across many devices without the data leaving any of them. The appeal here is plain: a system that gets better without your job records being pooled somewhere you cannot see.
Quantum hardware pointed at problems like routing and scheduling. The theory is strong; the machines are not there. It is on this list so the timescale is stated honestly rather than sold.
Systems that plan several steps and act without being prompted at each one. Fine when the steps are dull and reversible, dangerous the moment one of them books a truck. Scope narrowly or not at all.
Making a system show what it read and why it decided. Not an academic concern in this work: an extracted invoice figure nobody can trace is a figure nobody should file.
What I am building or pulling apart at the moment. The status label says how far along each one is, and most of them are not far.
One intake that takes the phone call, the photograph and the typed note about a job and files all three against it, instead of three systems each holding a third of the story.
Answering questions across several sites' records without those records being copied into one pile first.
Watching live data — loads, sensor readings, calls — and putting the exceptions in front of whoever is on the board, ranked, rather than waiting to be asked.
An extraction system for the photographed and scanned paperwork that field work produces, which reports what it could not read instead of guessing at it.
Two working tools, free and unguarded. Both run entirely in your browser and give you a usable answer without talking to anyone.
Twenty questions about your records, your systems, your people and the job itself. It scores each one and tells you which is actually holding you back.
Start the assessmentModel the cost, payback period and three-year return of an AI project against your own headcount and use case before committing budget to it.
Open the calculatorIf your problem has no obvious off-the-shelf answer, that is the kind I want to hear about. Bring the job everybody else told you was not worth automating.
Book a call