Foreman's hands-free headset

The problem

What explains a bad aggregate batch is in one person's head.

Critical operational knowledge (a hopper carrying more contaminants, a need to reinforce manual sorting) lives in the supervisor's head and never enters the system because they can't type while walking and managing the plant.

• The supervisor knows that this morning's hopper is carrying more plaster than usual, that the cabin needs a second picker on mixed loads, that the screen deck is blinding again.
• They spend the day walking between the picking cabin, the mobile plants and the landfill cell with gloves on and a radio in hand. Typing is not an option.
• So that knowledge never enters the system: it survives in a verbal handover and, with luck, a line in a notebook.
• When a recycled aggregate batch fails its grading or its contaminant count, nobody can link it to what the supervisor saw at nine in the morning.

How it fits the IRIS system

Connect in voice mode — the supervisor speaks, iLEAN structures and links.

Connect's voice mode with a Bluetooth headset and a custom wake word ('iLEAN, note this'). An LLM structures the dictation as an incident linked to the active hopper, batch or shift, and returns support by voice or tablet push within seconds.

It records what the supervisor chooses to tell it, when they choose to, after the wake word. That is what keeps the case alive months later: it is a tool for the supervisor, not a microphone pointed at the crew. And the incident lands where it matters: on the hopper, the batch or the shift it belongs to.

See the full IRIS architecture →

Before and after

The supervisor's round, lost versus captured

Aspect	Today	With iLEAN Connect
An incident at the hopper	Told on the radio and forgotten	Logged against the active hopper and batch
Recording it	Walk back to the office	Say “iLEAN, note this”
Reinforcing manual sorting	Decided and not recorded	Decision with time and reason
Link to scale and sorting data	None	Cross-checked live
Asking for a figure on the yard	A call to the office	Answered by voice or tablet push
Next shift's starting point	A five-minute chat	The shift's incidents, in order

Knowledge that used to die at shift's end → enters the system in real time and cross-checks live scale and sorting data.

Impact estimate

Impact estimate — an enabling piece, with no number of its own.

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. We do not fix a range, because its value depends on how many incidents die at shift's end today, and that is only measured once they start being recorded.
• Recycled aggregate quality depends on it: the plaster or wood the supervisor spots at the hopper explains the batch that later fails its contaminant count.
• Clean per-batch records depend on it too: incidents enter the system in real time instead of being reconstructed for an inspection.
• The figure is validated with the plant supervisor after the first weeks, counting how many incidents were captured that previously left no record.

Strategic value, hard to monetize upfront, but recycled aggregate quality and clean per-batch records depend on it. *Estimate to validate* with the plant supervisor.

And the fair question from the production manager

“Will it understand me next to a crusher running at full load?” — the headset filters the plant's noise at source, and turning a short dictation into an incident with hopper, batch and shift is an anchored task, where the best models drop below 1.5% error [1]. When a field is ambiguous, iLEAN asks back by voice instead of guessing, and the incident is confirmed on the tablet before it enters the batch record.

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