01
Manufacturing QC — surface defect detection, dimension checks, assembly verification.
Fits you
Edge inference live
97.3%
Detection accuracy
40ms
Inference latency
8
Factories live
Computer Vision
Your cameras are already watching. They just can't think yet.
CCTV is cheap. Production-line cameras are installed. Document scanners sit idle. What's missing is the model that turns pixels into decisions. We build that.
ManufacturingRetailBFSIHealthcareLogisticsAgricultureManufacturingRetailBFSIHealthcareLogisticsAgricultureManufacturingRetailBFSIHealthcareLogisticsAgriculture
Who this is for
Real CV use cases we've shipped.
If any of these sound like you, we should talk.
02
Retail — in-store analytics, shelf monitoring, footfall, demographic breakdown.
Fits you
03
Document intelligence — Aadhaar, PAN, GST, bank statements, forms-OCR with validation.
Fits you
04
Security / attendance — face recognition, helmet detection, PPE compliance.
Fits you
The proof
CV deployed at production scale.
97.3%
Defect detection accuracy
on FMCG line
< 40ms
Inference latency
on edge GPU
24/7
Continuous operation
no supervision
8
Factories live
across India
What we build
Computer vision systems we've shipped.
01
Defect detection
Missing / bent / cracked / misaligned.
02
Retail analytics
Footfall, heatmaps, dwell time.
03
OCR & extraction
ID docs, invoices, forms, signatures.
04
Face / object / pose
Attendance, PPE, safety compliance.
ClientA packaging manufacturerManufacturing
Zero-defect packaging line. <em>97.3% detection</em>, below 40ms inference.
Replaced manual QC (3 inspectors per shift) with a camera-based CV system. Runs on edge GPUs on-line. Detects missing labels, misalignment, seal defects.
01
97.3%
Detection accuracy
02
40ms
Inference latency
03
3
Inspectors reassigned
04
₹0L/mo
Customer returns saved
What it's like working with us
We had 2% customer-side rejection on packaging. Now it's 0.08%. The line didn't slow down.
RJ
Rakesh J.
Plant Head · A packaging manufacturer
Manufacturing · Gujarat
Tech stack
The tools. Chosen for your reasons.
09technologies in rotation
01PyTorch
02YOLOv8
03ONNX
04TensorRT
05OpenCV
06Triton
07NVIDIA Jetson
08Roboflow
09FiftyOne
Process
How we actually work.
06 stages
- 01
Feasibility
Weeks 1-2Can this be done with CV? Accuracy target?
- 02
Data collection
Weeks 2-4Cameras, lighting, sample set, edge cases.
- 03
Annotation
Weeks 3-6Labelled dataset, 2,000-20,000 images.
- 04
Model training
Weeks 6-9YOLO / custom, hyperparameter search, validation.
- 05
Edge deploy
Weeks 9-11Jetson / Coral / cloud. Optimise for inference time.
- 06
Production + retrain
Week 12+Monitor drift, capture edge cases, retrain monthly.
Questions
Answers, without the fluff.
Still have questions? Talk to us — we answer within a business day.
07common questions
01How much data do we need?
For defect detection: 2,000-5,000 labelled images per defect class, typically. We help design the collection strategy.
02Cloud or edge?
Factory lines: edge (Jetson). Retail analytics: cloud is fine. Privacy-sensitive (face): edge, always.
03What about reflective surfaces, dust, glare?
Lighting is 80% of production CV. We spec lighting rigs as part of the solution — you'll never get a good model on bad pixels.
04How long does a typical engagement take?
Most projects run 10-18 weeks from kickoff to production launch. We share a milestone plan in week one and update weekly.
05Do you sign an NDA?
Yes. Standard mutual NDA on request, before the first technical conversation.
06Who owns the code and IP?
You do. Code is in your GitHub org from day one. All IP transfers unambiguously on delivery.
07What does your pricing model look like?
For v1 builds: fixed scope, fixed milestones. For ongoing work: monthly retainer with a defined team. We don't do time-and-material surprise billing.
Related
Explore more of what we do.
Eyes that don't miss.
30-minute feasibility call. Send us sample images. We'll tell you what's achievable.
Free CV feasibility.