Industrial AI that starts from good.
Experiq AI builds machine intelligence for manufacturing. Our first product inspects fabric on a running line — learning from normal production, flagging what doesn't belong, and putting a person in the loop to make the call.
One product today. A platform underneath.
Fabric inspection is where we started, not where we stop. The same foundation — visual understanding, anomaly detection, edge deployment and a disciplined feedback loop — applies across any process where quality is judged by eye.
Visual understanding
Feature learning over images and video, from texture and surface defects to counting, measurement and presence checks on a moving line.
Anomaly detection
One-class methods that model what normal looks like and score everything else. Built for the real case: defects are rare, unlabelled, and nobody wants to label thousands.
Edge & on-premise
Inference on the factory floor with modest hardware. Deterministic latency, no dependency on the plant's internet link.
Human-in-the-loop
Every operator decision — defect, good, or undecided — is recorded and fed back, so the system improves from use rather than from a one-off training run.
Licence-clean stack
Built on permissively licensed foundations. No AGPL in anything we ship, so your legal team never has to have that conversation.
Custom model development
Where the problem needs something the platform doesn't cover, we build it — scoped, measured, and handed over with the numbers to judge it.
What we ship.
Our first product is in the field. Others are in development, and we'd rather say so plainly than pad a product page.
Experiq Inspect — fabric defect detection
A continuous inspection system for fabric on a running line. It learns the normal appearance of your cloth from defect-free production, scores every frame against that memory, and routes anything unusual to an operator for a decision — complete with the roll number and the metre position, so the fault can be found on the roll rather than hunted for.
Surface inspection, other materials
The same one-class approach applied beyond cloth — leather hides, plywood and veneer, film, sheet metal and ceramics. Different optics, same foundation.
Production analytics
Turning the inspection record into something a plant manager can act on — defect trends by shift, machine and operator, and where quality actually drifts.
Supervised defect classification
Once enough defects have been labelled on site, naming what the defect is becomes possible — the step from "something is wrong here" to "this is a broken pick".
Capture, detect, verify, learn.
A closed loop. The value isn't only in what the model finds — it's in what the plant teaches it afterwards.
Capture
An industrial camera watches the moving cloth. The line encoder gives every frame a position, so a defect is tied to a metre on a specific roll.
Detect
A one-class model compares each frame against a memory of normal cloth. Anything that doesn't fit is scored, and the likely region is marked on the image.
Verify
The operator sees the photo, the roll and the position, and answers: real defect, good cloth, or undecided. That judgement is recorded against their name.
Learn
A false alarm is folded back into the model, so the same pattern stops coming up. An undecided item comes back for someone who knows the cloth.
Where eyes are still doing the work.
Anywhere quality is judged visually on a moving web or a repeating part, and defects are too rare to label.
Built for the plant, not the demo.
No labelled-defect prerequisite
Supervised inspection needs thousands of labelled examples before it works. Most plants don't have them and can't afford to create them. Our approach starts from good production instead.
Numbers you can argue with
We report the false-alarm rate alongside the catch rate, and the operating point we chose to get there. A detector that catches everything by flagging everything is not a working detector.
Deployed, not just trained
The hard parts are the ones that come after the model: optics, lighting, throughput, the operator's screen, and the shift change. We build for those.
You own the outcome
Your images stay in your plant, your inspection record is yours to export, and the models we fit on your production are fit to your production — not a generic model wearing your logo.
Start with one line.
We'd rather prove it on your worst line than pitch it for your whole plant.
Walk the line
We look at the process, the defects that actually cost you money, and where a camera can realistically sit. No software until that's clear.
Pilot on one line
Camera, lighting and a workstation on a single machine. We calibrate on your good production and run it against your real output.
Measure, then scale
You get the catch rate, the false-alarm rate and the cost per inspection. If the numbers work, the second line takes a fraction of the effort.
What it takes to run.
| Camera | Industrial machine-vision camera suited to your web width and smallest defect of interest. Resolution is driven by the arithmetic — the smallest defect you care about needs enough pixels across it to exist at all. |
| Lighting | Purpose-built LED illumination. In surface inspection this matters more than the camera, and it is designed per material. |
| Compute | A single GPU workstation on the factory floor. One machine can serve several cameras. |
| Integration | Encoder input for position, and a screen for the operator. Optional link to your existing quality or ERP record. |
| Network | Runs on the plant LAN. No continuous internet connection required. |
| Calibration | A set of defect-free production images from the line, captured under production lighting. Typically a day. |
An applied AI company, starting in the mill.
What we are
Experiq AI is an applied artificial intelligence company based in [City], India. We build systems that do useful work in physical environments — where the data is messy, the hardware is fixed, and the answer has to arrive while the line is still running.
Why we started here
Fabric inspection is a problem where the technology finally meets the economics: defects are rare, the judgement is visual, the labour is skilled and scarce, and the cost of a missed defect is real. It's a good place to prove a platform — and the same foundation carries into every other surface, sheet and web.
Put a camera on your worst line.
Tell us what you make and which defect hurts most. We'll tell you honestly whether vision can catch it, what it would take, and what it would cost.