Defect detection
that runs where
it matters.
Chitti AI — Moving Manufacturing Beyond Jugaad.
On-edge computer vision for manufacturing QC. Detects defects in under 200ms, generates immutable audit records, and works when the network doesn't.
Supported By
Quality control is still running on paper and hope.
Three systemic failures that every plant manager knows but can't fix with existing tools.
Defects escape manual inspection
Visual fatigue is real. After hour six, operators miss scratches, porosity, and mislabels that ship directly to customers. The cost isn't just rework — it's reputation.
Paper records prove nothing
When a recall hits, you need visual evidence tied to every unit. Manual logs can't provide it. Without traceability, every defect becomes a potential liability.
Cloud AI breaks when the network does
Factory internet is unreliable by design. Cloud-first solutions create a single point of failure. When latency spikes or connectivity drops, the line stops.
See the inspection loop in action.
Camera feed flows through edge inference. Defects are classified in real-time. Every verdict generates an immutable audit record.
Built for the factory floor.
Edge inference under 200ms
ONNX models run locally on factory compute. No cloud round-trip for the critical detection path.
Offline-first
Inspection continues during network outages. Sync happens asynchronously.
Immutable audit chain
SHA-256 hash blocks for every verdict. Tamper-evident, court-ready.
Model router
Swap defect models without redeploying. Weld, paint, battery, packaging — all on one edge device.
Human override
Operators can override any verdict. Every override is logged as a first-class audit event.
Six-week pilot
One line, one defect class, one proof point. Scale only after the data justifies it.
WhatsApp alerts
Real-time defect notifications to floor supervisors. No dashboard login required.
Pilot reports
Auto-generated PDF audit reports. Compliance-ready, no manual compilation.
Cameras see what operators miss.
Industrial cameras capture every unit on the line at 30fps. No manual logs. No paper trails. Just continuous, timestamped visual evidence.
Models run locally. Decisions in 124ms.
ONNX models execute on edge hardware. Defects classified in under 200ms. The critical path never leaves the factory floor.
Sync when connected. Work when not.
Edge nodes queue audit records locally. When connectivity returns, they sync to the cloud. Dashboard updates. PDF reports generate automatically.
Every decision is provable.
Hash-chained audit records. Operator override trails. Timestamped visual evidence. Built for Schedule M and beyond.
Start with one line.
Prove one defect class.
Six-week pilot on a single production line. Scale only after the data proves it.
Begin Pilot