The knowledge that walks around and dies at end of shift

The problem

The shop's critical knowledge is oral and it walks.

The critical knowledge of a mold shop is oral and it walks. Nobody asks the supervisor anything in writing because they know he cannot type, and he records nothing because the cost of recording is interrupting what he is doing. There are three consequences and all of them show: the project lead does not know what he knows, the night shift gets second-hand instructions, and the day he is out the shop measurably slows down. This usually gets called "key person dependency" and accepted as inevitable. It is not: it is an interface problem.

• The supervisor spends the day between the fitting bench, the five-axis machines, EDM and the spotting press, drawing in one hand and caliper in the other. He makes dozens of calls a day and none of them get written down.
• Nobody asks him anything in writing because he cannot type, and he records nothing because recording means walking back to the office.
• The project lead does not know what he knows, the night shift gets second-hand instructions, and the day he is out the shop measurably slows down.
• This gets called key person dependency and accepted as inevitable. It is not: it is an interface problem.

How it fits the IRIS system

Connect in voice mode — iLEAN listens to the supervisor, not to the toolmakers.

Connect in voice mode over a Bluetooth earpiece — the hands-free mode of IRIS.

He dictates what he has just decided without stopping: this cavity needs weld repair and remachining, that slide insert moves to the night machine. iLEAN structures it against the mold and the detail, tells whoever needs to know, and hands support back in seconds — including the warning that the finishing program was already released and needs rebuilding. The call is always his; what the shop gains is that the call survives the end of his shift and reaches the toolmaker who has to execute it.

• He dictates what he has just decided while walking: this cavity needs weld repair and remachining, that slide insert moves to the night machine.
• iLEAN listens to what he directs at it, structures it (mold, detail, decision, who needs to know) and stores it against the right project.
• And it returns immediate support: "logged against that mold; the finishing program was already released, alerting programming to rebuild it with the weld repair included".
• The shift handover inherits written context instead of someone else's memory.

Humans in command, without qualification: the call is always his. The AI records, alerts and prepares the ground.

See the full IRIS architecture →

Before and after

Today's shop versus the captured shop

Aspect	Today	With iLEAN Connect
A bench decision	In one head	Logged against mold and detail
Shift handover	Memory of whoever is leaving	Written context for whoever arrives
Released program after a weld repair	Found out at the machine	Programming alerted in seconds
Who decided what and when	Nobody can say	Traceable
Supervisor's effort	Walk back to the office	Speak while walking
Ordinary floor conversation	—	Not recorded

from zero record of bench decisions → history per mold and per detail. From a handover based on the memory of whoever is leaving → context available for whoever arrives. From rework caused by lost instructions → traceability of who decided what and when.

Impact estimate

Impact estimate — to be validated with your 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.

• Estimated payback 4-10 months.
• Mostly from rework that stops happening: the instruction that got lost between shifts and was machined the wrong way.
• And from how much faster each shift starts, because the night crew inherits written context instead of second-hand orders.
• Humans in command, without qualification: the AI records, alerts and prepares the ground.

estimated payback 4-10 months, mostly from rework that stops happening and from how much faster each shift starts. *Estimate to validate*.

And the fair question from the production manager

"Is this for tracking my supervisor?" — no: iLEAN listens only to what he directs at it, when he chooses to. There is no continuous listening and no location tracking. Structuring a dictated note into mold, detail, decision and recipient is an anchored task, where the best models drop below 1.5% error [1], and what he gains is not having to remember at the end of the shift what he decided eight hours earlier.

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