01 / FEATURE
Defect detection at line speed
Surface defects, missing components, assembly errors caught before parts leave the line. Sub-100ms inference; every miss captured for retraining.
Operational intelligence.
Edge vision, predictive maintenance, and yield analytics — deployed on the floor, not in the cloud. Built for plants, warehouses and logistics networks across ASEAN.
The pressure
Why factory AI fails.
The cloud demo works. The floor pilot runs for two weeks. The system gets unplugged. Manufacturing AI fails when it ignores the constraints of the environment — latency, ruggedness, and the operator who has to live with it.
Latency the line can’t tolerate. A 200ms round trip to the cloud is a defect through the next station. Inference has to happen at the edge, on hardware that survives the floor.
Active learning, every shift. Defects evolve. New SKUs show up. A model trained on last quarter’s data drifts within weeks if there’s no capture-and-retrain loop running quietly in the background.
Yield and OEE locked in spreadsheets. Operations data scattered across MES, ERP, SCADA and shift logs — visible only after the shift, never in time to intervene during it.
Safety and compliance that won’t yield. PPE compliance, exclusion zones, near-miss detection — the system has to support safety, not annoy the operator into bypassing it.
Where we ship
Use cases we’ve put into production.
Patterns deployed across electronics, F&B, automotive and logistics in ASEAN — every one running on edge hardware with an active-learning loop back to the central training pipeline.
02 / FEATURE
Predictive maintenance
Vibration, temperature and acoustic signals fused with maintenance logs. Honest false-alarm rates — operators see the alerts they should act on, nothing more.
03 / FEATURE
Yield & OEE analytics
MES, ERP and SCADA streams unified into a single yield picture. Closed loops back into setpoint and changeover decisions during the shift, not after it.
04 / FEATURE
Safety-compliance vision
PPE detection, exclusion-zone monitoring, near-miss capture — instrumented for safety officers, not as a surveillance overlay on operators.
05 / FEATURE
Warehouse & logistics optimisation
Slotting, pick-path planning, dock-door scheduling — informed by real-time order flow and live yard visibility, not nightly batch runs.
06 / FEATURE
Document intelligence for ops
Invoices, customs forms, BOMs and trade docs extracted with field-level confidence. Layout-aware, multilingual, audit-ready.
Real-world example
Half a million dollars in scrap, recovered.
An electronics contract manufacturer was scrapping a five-figure number of boards per shift to a defect family their visual inspection couldn’t see. We deployed an edge vision system on three SMT lines, trained on the operator’s own escape data, with an active-learning loop that captured every override.
Before
Status quo
- Defect escape rate measured monthly, in arrears
- Cloud-only AOI pilot abandoned for latency
- No retraining loop — model frozen at deployment
- Operators bypassing alerts they didn’t trust
After
Post-deployment
- Defect escape rate visible per shift, per station
- Sub-100ms edge inference, no cloud round trip
- Weekly retraining on operator-captured edge cases
- Operators using the system to argue for line-rate changes
ASEAN electronics manufacturer · 6 months post-deployment
Solutions that fit
Where to start, by maturity.
AI Sprint — 4 weeks → Validate a defect or yield use case end-to-end with a working prototype.
Vision AI — defect → Edge-deployed defect detection at line speed with active-learning loops.
Accelerate — embedded → One senior engineer in your sprint cadence, 3–6 months, monthly cancel.
Deploy — production SLA → Two to three engineers, full FORGE, monitoring and SLA-backed support.
Compliance & assurance
Frameworks we build against.
Manufacturing systems live in tightly-regulated supply chains. Quality and safety evidence ships with the deployment, not in a binder six months later.
ISO 9001 / IATF 16949. Quality management evidence, traceability and process control — instrumented into the model lifecycle, ready for customer and notified-body audits.
Workplace safety. Aligned to local OSH frameworks across ASEAN. Vision systems support the safety officer; they don’t replace them and they don’t surveil the operator.
Data residency for IP. Process data, recipes and yield numbers are crown jewels. Tenant isolation by default; nothing leaves the plant perimeter without explicit policy.
Got a defect family your AOI keeps missing?
Talk to our manufacturing lead. We’ll review your escape data and recommend an edge-vision starting point.
Talk to manufacturing lead
30 minutes · reply within 1 business day