Vision models are accurate in the cloud; one inference on the line costs hundreds of milliseconds plus jitter. Takt cannot queue. Latency budget decides whether the model lives at the station or the data center. Edge releases; cloud trains, spot-checks, and versions. Wrong placement makes accuracy a stop-line source.
Where the model lives is not architecture taste—it is whether the conveyor can wait.
Write Latency Before Picking Place
Measure allowed time from photo to release. Over budget means edge deployment or smaller inputs. Offline edge still releases on last model and buffers. Cloud updates must roll back—no silent push at peak.
- Do not put online learning directly on the release path.
- Spot checks and hard cases may go cloud; takt pieces must stay local.
- Edge health checks—fans and disks maintained like production.
Edge Releases, Cloud Evolves
The XYN digital intelligence system writes QC results to work orders; inference placement follows station latency. In the right place, intelligence is part of takt; in the wrong place, intelligence fights takt.
Stopwatch photo to release. Over budget—edge first, bigger cloud model later.
