The model works on the test set. Then the lighting changes, the camera moves, and nobody owns it. We build the part that survives contact with a real site.
First scoping call is free. We'll tell you if it isn't feasible.
The hard part was never the architecture. It's everything between a promising result and a system somebody trusts on a Tuesday morning.
Registered in England & Wales.
Three stages. The first is deliberately cheap, because it's the one that tells you whether to continue.
We look at what you can actually capture, tell you whether the problem is solvable with it, and what accuracy is realistic. Including when the answer is no.
From labelling strategy through to a system running where it needs to run — your servers, or a device on site with no reliable network.
Accuracy is tracked against real outcomes, drift is flagged before anyone notices manually, and retraining is a routine rather than a crisis.
Forty-five minutes. You describe what you want the system to see and decide. We tell you whether it's a vision problem at all, and what it would realistically take.
We work with your existing footage or capture a sample, and come back with an achievable accuracy figure and what it depends on. Fixed fee, no commitment beyond it.
If the numbers justify it, we build and deploy. You get the model, the weights, the code and documentation your team can maintain.
Forty-five minutes with people who have put vision systems into production. We'll tell you if your problem is solvable, and if it isn't, we'll say so.
The scoping call is free and the two-week feasibility stage is a fixed fee quoted after it. We deliberately keep that stage small, because it's the one that tells you whether to spend anything else.
Not necessarily. Sometimes existing CCTV or process footage is enough. If it isn't, part of the feasibility work is telling you what to capture and how much of it.
Yes. We deploy to edge devices on site where the network can't be relied on, and design around the bandwidth you actually have rather than the bandwidth on the spec sheet.
That depends on your data and conditions, which is exactly what the feasibility stage answers. Anyone quoting a figure before seeing your images is guessing.
You do. Model, weights, training code and documentation are handed over. Nothing is locked to us, and nothing stops working if you stop paying us.
We tell you at the feasibility stage rather than the build stage, which is the whole reason that stage exists. You'd have a documented answer for the next time it's proposed.
Tell us what you need the system to see, and what decision it would drive. We'll come back with a slot, or an honest reason not to bother.