Visual inspection and quality-control AI fail when vendors optimize for demo accuracy instead of line-speed false-alarm rates. Use this RFP checklist before shortlisting a computer vision partner in India.
TechieYan Technologies deploys production vision systems for manufacturing, logistics, and defence-adjacent programmes from Hyderabad — with edge inference, PLC integration, and source-open delivery.
1. Define acceptance metrics upfront
Require sensitivity, specificity, and false-alarm rate at production line speed — not offline accuracy on a curated dataset.
2. Edge vs cloud inference
Ask where inference runs (Jetson, Coral, IPC, cloud) and maximum latency at peak throughput. See our computer vision practice.
3. Lighting and camera variability
Vendor must document calibration under your plant lighting, vibration, and SKU change scenarios.
4. PLC and reject mechanism integration
Vision without reject logic is a dashboard. Require Modbus, OPC UA, or digital I/O integration in scope.
5. IP and model ownership
You should own weights, training pipelines, and deployment scripts — with NDAs signed before data sharing.
6. Pilot structure
Fixed-scope pilot on one line or station with go/no-go criteria before plant-wide rollout.
Related resources
FAQ
What should a computer vision RFP include?
Line-speed metrics, edge deployment target, PLC integration, lighting variability tests, IP ownership, and a fixed-scope pilot with documented acceptance criteria.
