The UHT line startup sheet, digitized with one photo

The problem

The two documents that legally define the batch spend the morning in a folder.

The sheet is printed first thing with ERP data, and what actually happens gets written on top of it: the silo change, the temperature adjustment on the sterilizer, the malformed pack ten minutes into the run. The CIP release record is handwritten by standard. These two documents are what legally define the batch. And both end up in a folder in the office, where they can sit for four to eight hours before anyone digitizes them. That gap goes unnoticed while everything runs well. It gets noticed on the day the destination market's health authority arrives for re-inspection and asks for evidence that this specific batch started on a clean circuit. Then someone has to go and find the paper, and the paper may not be where it should be.

• The startup sheet prints first thing with ERP data — product, format, batch, source silo — and what actually happens gets written over it by hand: the silo change, the sterilizer temperature adjustment, the malformed pack ten minutes into the run.
• The CIP release record is handwritten by standard. The CIP operator signs that the circuit was left clean, and that signature is the only proof the batch started on a clean line.
• Both go into a folder in the production office and wait there four to eight hours, sometimes until the next morning, before anyone types them into anything.
• During that gap the batch exists physically on the pallet but in no queryable system — and that is exactly the window in which a re-inspection or a customer complaint can ask for it.

How it fits the IRIS system

Connect in photo mode — the operator keeps the sheet and adds two photos.

Connect in "photo of the analog" mode. Step by step:

It is the simplest of the twelve to deploy and the one that meets least resistance at the filler, because it asks for nothing new: the operator fills in the sheet as always and photographs it together with the CIP record. The paper still exists; it just stops being the only place the data lives. That is usually why it is the first case a plant approves.

• The operator starts the line exactly as always and fills in the sheet by hand.
• When finished, they take two photos with a mobile or industrial tablet: the startup sheet and the CIP release record.
• A language model grounded in the plant's own vocabulary extracts the structured fields: batch, product, format, source silo, sterilization temperature and hold time, release signature, handwritten notes.
• Everything is stored in iLEAN's central memory against the batch number, along with the original image of both documents.
• A query by batch returns the startup sheet, the CIP record and the photo of the first pack on a single screen.

No habit change, no new form and no tablet the operator has to learn. The paper still exists; what changes is that it stops being the only place the data lives.

See the full IRIS architecture →

Before and after

Today's startup versus the captured startup

Aspect	Today, on paper	With iLEAN Connect
Latency until the batch exists digitally	4-8 h, more across a night shift	Second zero
Sterilization temperature and hold time	Handwritten on the sheet	Extracted and tied to the batch
CIP release signature	In a folder, hopefully	Photographed and linked to the batch
Proving a clean-circuit startup at re-inspection	Somebody goes to find the paper	One query by batch number
Handwritten deviations	Lost unless someone types them	Read and stored with the original image
Operator's routine	—	Unchanged: two photos

from 4-8 h of latency between startup and digital record, to second zero. From half a morning searching last month's binder, to the time it takes to type a batch number. From evidence scattered on paper, to evidence queryable by batch with the original image attached.

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.

• Estimated payback 4-9 months, depending on the number of line startups per day.
• Counting only the transcription time recovered by the production manager and the quality clerk, who today retype what the operator already wrote.
• From 4-8 hours of latency between startup and digital record, down to second zero — and from half a morning searching last month's binder, to the time it takes to type a batch number.
• Not counted, and larger: having the startup sheet and the CIP release ready on the day the destination market's health authority arrives for re-inspection.

estimated payback 4 to 9 months depending on startups per day, counting only the transcription time recovered by the production manager and the quality clerk. *Estimate to validate.* It excludes the value of having evidence ready on audit day, which is the larger and harder-to-quantify argument.

And the fair question from the production manager

«What if it misreads the sterilization temperature on the sheet?» — reading fields off a document whose structure is known is an anchored task, where the best models drop below 1.5% error [1]. And there is a second net: every value is checked against its physical range — a UHT hold temperature has a narrow window — and any low-confidence field is flagged for the quality clerk to confirm before it is written against the batch.

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