Hands-free on site: the foreman dictates, iLEAN logs it

The problem

The best-informed person on the site is the one who cannot write anything down.

Critical operational knowledge for each work zone — a machine sounding off, an operator needing backup, a stockpile about to fill up — lives only in the foreman's head and never enters the system, site after site. No one asks them anything because they can't type with both hands busy while walking.

• The demolition foreman spends the shift walking between work zones, directing machines by radio and by hand. Both hands are busy, the helmet does not come off and there is no desk anywhere in sight.
• In their head is everything that matters: a breaker sounding wrong since the morning, an operator who needs a second pair of eyes before dropping a wall, a stockpile about to fill and block the loader's run to the crusher.
• Nobody asks them for any of it, because asking means asking them to stop, take off a glove, take out a phone and type. So it gets solved on the spot and is never recorded anywhere.
• The consequence shows up later in the owned fleet: the same breaker fails on the next site, the repair is charged to that job, and nobody can say it had been sounding wrong for three weeks beforehand.

How it fits the IRIS system

Connect in voice mode — the foreman dictates, iLEAN structures and answers.

Connect's voice mode with a custom wake-word ("iLEAN, note this"). A grounded LLM structures the dictation as an issue linked to the active zone, machine and shift, and replies with useful voice feedback seconds later.

It listens to the foreman, not to the crew, and only when the wake word is spoken. That distinction is not cosmetic: it is what keeps the earpiece in use in month six instead of being left charging in the site office. On a site where subcontractors come and go by the week, anything that feels like being tracked is dead within a fortnight, and it takes the rest of the project down with it.

See the full IRIS architecture →

Before and after

Today's shift versus the captured shift

Aspect	Today	With iLEAN Connect
What survives the shift	Whatever someone remembers	The issue, tied to zone and machine
Cost of recording something	Stop, take out a phone, type	Say it while walking
Reply to what was dictated	None	Voice confirmation in seconds
Looking up a figure in the field	Walk back to the site office	Ask and hear the answer
Fleet history	Starts at the breakdown	Starts at the first symptom
Handover to the next shift	Five verbal minutes	A written summary that already exists

Before: operational knowledge dies at shift's end. After: it enters the system in real time and is cross-referenced with live maintenance and planning data.

Impact estimate

Impact estimate — an enabling piece, and it is worth saying so.

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.

• The brief is explicit and so are we: this case is hard to monetize directly and carries no payback of its own.
• Its value is enabling: it is the foundation of any preventive-maintenance plan for the owned fleet, which today starts the day a machine stops.
• Without it, the symptoms a foreman notices weeks in advance never become a maintenance record, and the fleet keeps being repaired after it stops.
• It is a case to size with your fleet manager, not with an invented figure that would not survive a management meeting.

Hard to monetize directly, but the foundation of any preventive-maintenance plan for the owned fleet. Estimate to validate with the fleet manager.

And the fair question from the production manager

"Our foremen talk in site slang, over a breaker." — that helps rather than hurts. Transcribing speech over a closed, known vocabulary — zones, machines, attachments, symptoms — is an anchored task, where the best models drop below 1.5% error [1]. What gets structured are those entities, not free prose, and anything dictated with low confidence comes back for confirmation in the same earpiece.

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