Easarc Vision — AI quality inspection
Catch the defect on the line, not in the buyer's warehouse.
Mount an ordinary IP camera. Vision flags fabric defects, print misalignment and packing errors in real time, and tells you which shift and which machine they came from.
The inspection view
A defect, the moment it happens, with its address attached
Knowing that 2.4% of the lot is defective is not useful. Knowing that it started at 14:20 on machine PR-04 after the shift changed is what lets you go and fix it.
2 defects in the last 60 seconds · linked to ESR-2026-0431
Defect mix, last 24h
- Weft bar34
- Print misalignment28
- Oil stain21
- Shade variation18
- Hole / cut12
By shift
| Shift | Output | Defect |
|---|---|---|
| A | 41,200 m | 1.8% |
| B | 38,600 m | 2.4% |
| C | 27,400 m | 3.6% |
Shift C is at 3.6% against your 3.0% limit. Supervisor notified on WhatsApp at 06:12.
- A
- Every frame carries the camera, the machine and the shift. That is what turns a defect count into something a production head can act on.
- B
- Defects attach to the work order running on that machine at that minute, so the QC report for a buyer builds itself.
- C
- The taxonomy is yours. A process house and a garment unit do not have the same defects, and a shared list helps neither.
- D
- Rates broken out by shift are usually the first uncomfortable thing the system shows you, and the first thing worth fixing.
How the pipeline runs
Capture at the edge, infer in Mumbai
Cameras push frames to a small edge box on your LAN. It buffers, samples and forwards — so a dropped internet link delays analysis instead of losing footage.
Capture
Any ONVIF or RTSP camera. Frames are sampled at the rate your line speed needs, not at whatever the camera happens to emit.
Buffer at the edge
AWS IoT Greengrass on a small box in your unit holds frames when the link drops, then drains the backlog when it returns.
Infer in the cloud
A SageMaker endpoint in ap-south-1 runs your tuned model. Round trip is typically under two seconds from frame to flag.
Act
The defect lands on the supervisor’s phone, on the work order in Flow, and in the daily QC report your buyer will ask for.
What you get
Everything Vision does
- Custom defect taxonomy per customer
- Edge capture with cloud inference
- Shift, operator and machine analytics
- Alerts when defect rate crosses your threshold
- Exportable QC reports for buyer audits
Integrations
It has to work with what you already run
Nobody replaces their whole stack for one vendor. These are live today, and the API covers the rest.
- ONVIF / RTSP cameras
- Any IP camera that speaks RTSP — no proprietary hardware
- AWS IoT Greengrass
- Edge capture and buffering when the line drops
- Amazon SageMaker
- Per-customer model training and tuning
- Easarc Flow
- Defects attach to the work order they came from
- Slack / WhatsApp
- Threshold alerts to the supervisor on duty
- S3 export
- Clip and report archive in your own bucket
Pricing
Easarc Vision plans
Every paid plan starts with a 14-day free trial and no card. Annual billing gives you two months free.
Starter
₹4,999/camera/mo
Billed monthly
- 1 camera
- Standard defect models
- 30-day clip retention
Growth
Most chosen₹12,999/mo
Billed monthly
- Up to 5 cameras
- Custom defect taxonomy
- Shift analytics
- 90-day retention
GST extra at 18%. All paid plans include a 14-day free trial, no card required.
USD shown for reference at ₹88 = $1. Invoices are issued in Indian rupees.
Get started
Put Vision in front of one line
You do not have to commit the whole unit. Run Easarc Vision against a single product line for a fortnight and judge it on what changes.
14-day free trial, no card required. GST extra at 18%.