Talk to the AI without stopping your walk

The problem

The person who knows the line best never has a free hand to record it.

critical operational knowledge lives in the line lead's head and never enters the system while it's still useful.

• The forming and welding line lead walks between stations the whole shift, hands busy: adjusting a stand, checking a seam, supporting a welder in training.
• They notice what no system records: a steel lot throwing more rejects than usual, a forming roll starting to wear, a seam showing porosity on one program.
• Recording it means walking back to a terminal, so it gets told at the shift handover — or not at all.
• When that knowledge dies at shift end, the next lead meets the same roll wear or the same steel lot from scratch, and quality hears about it only when a reject appears. The same problem gets solved again, by a different person, with no memory of the last attempt.

How it fits the IRIS system

Connect in voice mode — the lead speaks, iLEAN structures, nobody types.

Connect voice mode with a custom wake phrase. An LLM structures the dictation as an issue linked to the active lot/line/shift, and returns confirmation by voice or push.

A wake phrase, a sentence spoken while walking, and a confirmation back by voice or push. What the lead knows reaches quality while it can still change a decision, tied to the lot and the line it concerns. The shift handover stops depending on the memory of whoever is leaving.

See the full IRIS architecture →

Before and after

The shift in the lead's head versus the shift on record

Aspect	Today	With iLEAN Connect
A steel lot giving more rejects	Mentioned at handover, maybe	An issue tied to the heat and the line
Forming roll starting to wear	Noticed, then forgotten	Recorded with date and stand
A welder in training needing support	Depends on who is nearby	Logged and routed to the right person
Reaching quality	Next day, if at all	In real time
The lead's hands	Busy — so nothing is written	Still busy — the voice does it
Continuous improvement input	Anecdotes	Issues crossed with program and steel-lot data

knowledge that dies at shift end → enters the system in real time and gets cross-referenced with live program and steel-lot data.

Impact estimate

Estimated impact — to validate with your own numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• No standalone payback: this is an enabling piece. It is hard to monetize upfront, and we would rather say so than put a number on it.
• Its value is that every continuous-improvement effort on the floor depends on this information existing and reaching quality in time.
• Knowledge that died at shift end enters the system in real time, crossed with live program and steel-lot data.
• We estimate it together with the area lead, looking at the issues that reached quality too late last year.

hard to monetize upfront, but every continuous-improvement effort on the floor depends on this information existing and reaching quality in time. *Estimate to validate with the area lead*.

And the fair question from the production manager

“Is this for keeping track of what my people say?” — no. iLEAN listens to the line lead, and only after the wake phrase; there is no continuous listening and no location tracking. Turning a dictated sentence into an issue with line, lot and symptom is an anchored task with a closed trade vocabulary, where the best models drop below 1.5% error [1], and the lead hears the confirmation before it is saved.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.