Computer vision for UK manufacturers.
The defect that slipped through last Tuesday cost you a returned batch, an irritated customer, and three hours of root-cause meetings. The camera over the line could have caught it the moment it happened. I build vision systems for QA, count and verify, and exclusion-zone safety, running on the cameras above your line, on your own hardware, with the code in your repo.
Or copy: [email protected]
What's actually happening on your line right now.
Your QA person checks one in fifty parts, on a clipboard. Most of the time that's fine. Sometimes it isn't, and you find out at the customer end.
You've looked at vision-QA systems before. The quotes came back from integrators selling proprietary hardware and a five-year service contract, at a number that didn't work for one line, let alone the four you run.
There's a middle option that doesn't get pitched as often: a fixed-price build, trained on your actual parts, running on a small computer next to the line. Same accuracy. A tenth of the cost. The code is yours.
What I build for manufacturers.
Visual QA / defect spotter
Camera over the line, flags out-of-spec parts in real time. Trained on your actual product, your actual defects, in your actual lighting. Connects to a screen, a light, a buzzer, or your existing MES. Your choice.
Quoted on scope.
Fixed price in writing before any work starts. Hardware, on-site installation, and any opted-in subscriptions quoted separately at cost.
Count, sort & verify
Counts items in a tray, batch, or pallet. Reads labels and matches them to a manifest. Particularly useful at packing, dispatch, and goods-in.
Quoted on scope.
Fixed price in writing before any work starts. Hardware, on-site installation, and any opted-in subscriptions quoted separately at cost.
PPE & exclusion-zone alerts
Factory-floor version of the construction kit. Flags hi-vis breaches, hard-hat zones, forklift exclusion areas, machinery interlocks. Audit log for your H&S team.
Quoted on scope.
Fixed price in writing before any work starts. Hardware, on-site installation, and any opted-in subscriptions quoted separately at cost.
Cycle-time measurement
Tracks how long each step on the line actually takes, from camera footage. The data your continuous-improvement team has been trying to get manually for years, captured automatically.
Quoted on scope.
Fixed price in writing before any work starts. Hardware, on-site installation, and any opted-in subscriptions quoted separately at cost.
A few of the situations this system handles.
A production line where a small percentage of out-of-spec parts make it through manual QA. The system watches the line at a chosen station, flags suspect parts in real time using a light or a buzzer, and logs every flagged part with an image for the QA lead to review.
A packing area where items are loaded into trays or onto pallets and labelled. The system counts items, reads labels, and confirms what's loaded matches what was supposed to be loaded. Mismatches halt the line or flag for review.
A factory floor with exclusion zones around machinery and forklift routes. The system watches the zones, alerts when someone on foot enters an active forklift lane or steps inside a machinery exclusion zone, and logs every alert for the safety review.
A continuous improvement team trying to measure cycle time on a line without manual stopwatching. The system reads the line, logs the time each step takes, and produces weekly summaries the team can act on.
Why this, why me.
Trained on your actual parts.
Generic defect detection doesn't work. Every build starts with images from your line, in your lighting, of your products. The system gets better as you feed it more examples.
It runs on your line, not someone else's cloud.
The camera, the computer, the data: all on-site. No internet dependency for the QA decision. No cloud bill.
The code is yours.
Most vision-QA vendors lock the trained model and the dashboard. Walk away, lose everything. Mine ships as code, weights, and training notebooks. You can retrain it yourself when your product changes.
One person, start to finish.
I do the camera spec, the data collection, the training, the integration. No subcontractors, no rotating teams.
Fixed price, in writing.
The quote is the price. Re-training when your product line changes is a separate fixed-price job, with no monthly fees in the meantime.
A call, then a build.
We talk
Thirty minutes for manufacturing scoping. You walk me through the line, the defects, the volumes. I tell you what's achievable and what isn't.
I build
Three to six weeks, including the data collection phase. The longest part is gathering enough examples of real defects, and your QA team helps with that in week one.
Every quote breaks out the build fee, hardware, install labour, and any subscriptions in writing before you commit. One total, every line visible.
The things people ask before we start.
How accurate is it?
Depends entirely on the defect type, the lighting, and the line speed. The honest range for typical light-manufacturing visual QA is 95 to 99 percent on defects the system has seen examples of. Novel defects, ones it has never been shown, are handled by being flagged as uncertain rather than missed. We work through your specific parts and your acceptable false-positive rate on the call.
Can it learn new defects?
Yes, but it needs examples. The system handles things it has been shown reliably. Genuinely new defect types need to be added to the training set. I show you how to do this yourself as part of the handover, so you're not dependent on me every time the line throws up something new.
Will it slow down my line?
No. Detection happens in milliseconds on the camera-side computer. The line keeps moving at its normal speed. The system reads the line; it doesn't sit in the middle of it.
What about my existing MES or ERP?
Most modern systems have an API or accept a webhook. Detections can land in your existing tooling, or the system can stand alone with a screen and a buzzer at the line. We pick the right level of integration based on what's actually useful, not what's possible.
Do I need to buy a special camera?
Usually no. Most modern IP cameras work fine for QA at typical line speeds. Very fast lines or very small defects sometimes need a specific industrial camera. I tell you which camp you're in during scoping.
What happens when I change product line?
The system needs retraining for new products. Small changes (different colour, different size of the same part) often work without retraining. Significant changes need new training images and a retraining job. This is a fixed-price piece of work, typically 800 to 1,500 pounds depending on how different the new product is. Or you can do it yourself using the training notebooks I hand over.
How quickly can it be running on my line?
Three to six weeks for the first line, including the data collection phase. The longest part is gathering enough examples of real defects, which your QA team helps with in week one. Subsequent lines run faster, typically two to three weeks each, because the data pipeline and configuration are already proven.
Tell me what you need built.
Email me a sentence about your line. I reply the same day.
Or copy: [email protected]