Platform

A camera, a model, a panel — and nothing on the line you don't already have.

NumTech is a per-line visual oversight system. It connects to the IP cameras you already own, runs a model trained on your own footage, and turns what it sees into structured events that an operator alert and a shift report both consume.

Request a demo

Real-time response

From frame to action in seconds.

Every feed runs through the NumTech AI worker. The instant an event appears it's flagged — and the response fires automatically: a stop, an alert, an escalation, before the next unit moves.

NumTech AI Worker

Buffer nearing overflow

Line 2 buffer area · 14:15:40

Line leader notified

Escalation sent · 14:15:42

Model lifecycle

Trained on your line, refined every shift.

01
Capture
We record a representative week of footage from the watched positions.
02
Label
Annotators under NDA tag the failure modes specific to your product mix.
03
Train
A line-specific model converges on the failure set in days, not months.
04
Deploy
The model runs on the edge appliance; updates ship without stopping the line.
05
Refine
False positives and missed events feed back into the next checkpoint.

Data residency

Footage stays in the country it was recorded in.

NumTech is hosted in Uzbekistan. Raw video stays on the line's edge appliance and is not transmitted off-site by default. The dashboard sees structured events — timestamps, types, confidence. You can opt into hosted storage of short event clips for review, and you keep the bytes either way.

Security & privacy

Built for a regulated factory floor.

01
Role-based access
Operator, supervisor, plant manager, and read-only roles. Audit log on every action.
02
No PII in training data
Faces are not part of the failure-mode training set. Worker visibility is not the product.
03
Encrypted in transit
Streams between the edge appliance and the dashboard are TLS-encrypted.
04
Local-first inference
The model runs at the edge; the line operates independently of the WAN.

FAQ

Questions we've already heard.

Do we need to replace the cameras we already have?

No. Any RTSP-capable IP camera on the line works. We connect to what's already pointed at the watched position; if a position isn't covered we'll tell you exactly where to add one.

Where does our footage live?

On the edge appliance in your network closet. It doesn't leave the site by default. You can opt in to hosted storage of short clips around an event for review, and you keep the bytes either way.

How long does setup take?

From signed paperwork to first events: a week of footage capture, two to three weeks of labelling and training, then deployment. Six weeks is the typical first go-live; lines we've seen before are faster.

What happens if the internet goes down?

The line keeps running. Inference happens locally on the edge appliance. The dashboard catches up on the event backlog when connectivity returns; operator alerts on the local panel are not internet-dependent.

Can it run in a noisy or dusty environment?

The cameras already on your line have to. The edge appliance is a fanless industrial box rated for the cabinet temperatures and conditions you're already running.

What if the model flags something that isn't a defect?

Operators can mark a flagged event as a false positive in one click. Those marks feed the next refinement cycle. We don't ship a model into a line we haven't tuned to that line.

Industries