The assembly supervisor dictates while walking, and iLEAN listens

The problem

Nobody asks the assembly supervisor anything — because they can't type.

In a Tier 1 premium LED headlight plant, the final assembly supervisor is not sitting in front of a screen: they are walking between assembly cells, hands full, seven hours a day. They are the first person to see the patterns no sensor reports:

• A drift in a process parameter — a cell that starts behaving differently long before the data trips any formal alarm.
• A correlation between an LED reel and rework — «since that reel came in, this cell reworks more» — that nobody logs in the moment today because there is nowhere to type between cells.
• The founded feeling that cell X is running worse this week — real operational knowledge that gets reconstructed days later, from memory, if it gets reconstructed at all.

All of that dies at the end of the shift. It never enters the system, never gets cross-checked against the line data and, above all, never feeds the program's live FMEA or the continuous-improvement plan: the occurrence and detection ratings get reviewed with whatever someone manages to remember in the weekly meeting, not with what the supervisor saw on the floor while it was happening.

How it fits the IRIS system

Connect in hands-free voice mode — the supervisor takes notes by dictating, without putting anything down.

The earpiece is not a system that records the supervisor during the shift: it is a tool they activate voluntarily when they want to note something down. They decide when they speak and what they dictate. Connect structures it, links it to the active plant context and returns it in seconds.

The supervisor says «iLEAN, note this». They dictate the observation in their own words, walking between cells. Connect structures it as a nonconformity candidate linked to the active program, cell, shift and operator. Seconds later, the response arrives by voice to the earpiece or as a push to the lead's tablet.

How Connect operates in earpiece mode in a premium LED headlight plant:

• Custom wake-word — the supervisor opens the note with a phrase of their own that only they say deliberately. Outside that phrase, the earpiece transcribes and sends nothing.
• Free dictation, no forms — they speak in their own words, just as they would tell a colleague: «cell 4 is giving more lens-bonding rework since the reel change».
• The LLM structures the note — Connect turns the dictation into a nonconformity candidate with clear fields, automatically linked to the program, cell, shift and operator active at that moment, without the supervisor having to recite any of it.
• Response by voice or push — the confirmation arrives at the same earpiece or, if more context is needed, as a push notification to the cell lead's tablet. The supervisor keeps walking, with the observation already in the system.
• Human verification before signing — nothing signs off as a formal nonconformity without passing through human verification on the tablet: a person reviews the draft, completes it and escalates it per the plant's quality-system procedure.

See the full IRIS architecture →

Before and after

Knowledge that dies at shift end vs. knowledge that enters the system in real time

Aspect	Without earpiece	With iLEAN Connect in earpiece mode
Operational knowledge from the shift	Dies at shift end, in the supervisor's head	Enters the system in real time and feeds the live FMEA
Taking a note between cells	No keyboard, hands full — nothing gets logged	Dictated by voice, hands free, without stopping walking
Pattern spotted on the line (drift, reel ↔ rework)	Reconstructed days later from memory, if at all	Captured the first time the supervisor sees it, structured
Link to program, cell, shift and operator	Depends on someone remembering and reconstructing it	Automatic — the system already knows what is active
Logging as a nonconformity	Written up hours later from memory, or not at all	Enters as a nonconformity candidate, pending human verification
Response to the supervisor	Nobody replies — the observation just disappears	By voice to the earpiece or push to the lead's tablet, in seconds

Impact estimate

Strategic value for your program — to be validated with continuous improvement.

This case has a different character from other Connect projects: its value is strategic, not directly monetizable. It doesn't replace a manual task with a calculable minutes saving; it is the enabling piece of the program's entire continuous-improvement system: it stops the assembly supervisor's operational knowledge from being lost. We frame it this way so the committee evaluates it with the right yardstick.

• Tier 1 premium LED headlight plant with a final assembly supervisor who walks most of the shift between cells, hands full and with no practical keyboard access.
• Connect pilot in earpiece mode on a single cell or shift — without changing the supervisor's role, without adding administrative tasks.
• The main return is not an accountable hours saving: it is feeding the live FMEA and the continuous-improvement plan with real data instead of memory reconstructions in the weekly meeting.
• There is an expected operational return too: the reduction of rework through early drift detection — the pattern the supervisor sees days before the rework piles up. That reduction is an estimate to be validated with the plant's continuous-improvement lead, not a generic figure invented here.

And the plant manager's fair concern

«Doesn't this end up being a system that records everything my supervisor says?» — no. It is exactly the opposite: the earpiece hears nothing until the supervisor activates it with their wake-word. There is no background recording, no continuous transcription, nobody listening to their conversation with a cell operator. The only thing that enters the system is what they themselves decide to dictate — because they want it on record, not because someone above asked for it. And the second concern, «what if the AI makes up half a nonconformity?»: for anchored tasks like this one — structuring a dictation against the active program, cell and shift — the AI has a reliability far from the hallucination problem of free-form generation ^[1], and even so nothing signs off as a formal nonconformity without passing through human verification on the tablet.

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