RIX: Real-time Intelligence eXpert

The Problem RIX Is Aimed At

Manufacturing plants lose critical knowledge when experienced workers leave, and roughly a quarter of the manufacturing workforce is expected to retire within five years. The knowledge that goes with them is rarely written down anywhere:

"Joe knows why Machine 5 acts up on humid days." "Maria always checks the hydraulic pressure first on that press." Neither of those is in a manual, and neither survives a retirement party.

ProAlert already records what happened: every work order, every part consumed, every downtime call, every scrap event, on one database. RIX is the layer that would read that record and turn it into a suggestion at the moment somebody needs it. That ordering matters... the intelligence layer is only possible because the operational record exists first.

That record is not hypothetical, and it is the part of this page that is shipping. Work procedures are versioned, and every work order pins the revision it ran. A typed execution log captures each transition: started, paused, measurement taken, step skipped with a stated reason, escalated, finished. A context snapshot records the shift, the operator, the asset state and the environment as they were when the order was raised. A technician who skips a step or adds one that was not on the template is recorded doing it. Whatever RIX eventually learns from, it will be learning from that, and it is being captured now.

Design Principles

Planned Capabilities

Work order decision trees

Raising a good work order takes tribal knowledge: which skills are needed, which parts are usually required, how long it should take, what safety precautions apply. RIX would guide the person through it. A technician reporting spindle vibration on a CNC mill gets asked whether it is constant or intermittent; constant points at spindle bearings and loads one template, intermittent points at tool holder balance and loads another. When it started and at what speeds it occurs narrow it further, separating bearing resonance from a mounting alignment problem.

Trees that learn from outcomes

The point is that the trees would not be static. On completion RIX captures the parts actually used, the actual time and the actual root cause, compares them against what it predicted, and proposes a change when a pattern emerges: "In the last 30 days, 8 of 10 spindle vibration work orders were actually caused by coolant contamination, not bearings. Should I add a coolant check to the tree?" A traditional decision tree is written once by an engineer and goes stale quietly.

Tribal knowledge capture

Four routes into the same store: pattern analysis across work orders, so that "senior techs resolve this three times faster" becomes a question about what they do differently; annotation, where an experienced worker adds "on this machine, also check the grounding"; plain language capture, where somebody tells RIX that a clicking hydraulic press is usually the relief valve rather than the pump; and outcome tracking, so the store is corrected by what actually worked.

Real-time pattern detection

Status and Honest Answers