Integrating the Jev AI Decision Model with Projection Guidance in Manufacturing: A Human-Supervised Proof of Concept for Bin Preparation

Source abstract

During part returns in manufacturing, deciding when to prepare an empty bin involves the current bin state, the remaining workload and the time needed for preparation. This study presents a human-supervised proof of concept linking projection guidance technology with Jev, a TypeSafe AI model that returns structured choices and uncertainty information. An existing picking workflow supplies the platform context. A separate extension converts depth-tool measurements into bin states, combines them with reported workload and preparation time, and asks Jev whether an empty bin should be prepared. A companion checks the response, records approval and projects request status. Four diagnostic calls to Jev were recorded across three software versions, all below a fixed confidence threshold of 0.80. In the final attempt, operator approval of a 0.75-confidence proposal created a preparation request; a later full reference triggered deterministic exchange handling. Local acknowledgments and confirmed emptying of the same bin completed one Bin1 lifecycle. The result demonstrates technical integration and decision traceability. Readiness was simulated; physical spare-bin exchange and external dispatch were not demonstrated. Zone events do not verify part identity, quantity or removal, and mask readings do not measure fill volume. Detection reliability and operational benefit remain unmeasured.