The fermentation curve, digitized without touching the equipment

The problem

The fermentation curve dies every time the screen rotates to another tank.

Fermentation temperature control shapes the final wine's aroma profile and color. Today that data dies on screen the moment it rotates to another tank, with no record linked to the batch.

• Temperature during alcoholic fermentation shapes the aroma profile and color of the finished wine. It is one of the few levers the winemaker truly controls, tank by tank.
• Yet the panel that governs it is often more than ten years old and isolated from the network. It shows one tank, then the next, and keeps nothing.
• What survives is a reading written down twice a day on a sheet hanging from the tank — a couple of dots where there should be a continuous curve, and nothing at all from the night shift.
• So when a batch comes out with a different profile from the year before, no one can tell whether a temperature peak caused it, in which tank, or on which day of fermentation it happened.

How it fits the IRIS system

Connect in photo mode on the panel — a camera in front of it, nothing inside it.

Connect photo mode on the panel. Fixed industrial camera, OCR plus a grounded LLM interprets each frame and structures the time-series curve per tank, cross-referenced with the harvest log (which batch is fermenting in which tank).

Replacing a fermentation control system during harvest is not an option, and it is rarely justified outside it. Reading it from the outside gives the winery the curve this campaign, and lets the replacement decision wait until there is a real reason rather than a missing record. Many wineries use that time to choose the right replacement calmly, with two campaigns of curves in hand.

See the full IRIS architecture →

Before and after

The panel on its own versus the panel being read

Aspect	Today	With iLEAN Connect
Temperature record per tank	Two handwritten readings a day	Continuous curve per tank
Link between curve and batch	None	Crossed with the harvest intake log
A temperature peak at night	Noticed only if someone was there	Recorded with time and tank
Explaining a profile change	Hypotheses	Root cause by batch and campaign in minutes
The control system	—	Untouched, nothing installed on it
Panel replacement	Pressure to bring it forward	Deferred until it is justified

zero correlation between curve and final profile → root-cause analysis by batch and campaign in minutes.

Impact estimate

Impact estimate — to be checked against your fermentation records.

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 5-10 months.
• Avoided premature replacement of the control panel: CAPEX that does not have to be spent this year just to get a record. The camera and its setup cost a fraction of a new control system.
• Better reproducibility of the wine's profile from one campaign to the next, because the curve that produced the good vintage is on record.
• And root-cause analysis by batch and campaign goes from impossible to a few minutes of querying.

estimated payback of 5-10 months, avoided premature replacement of the control panel (CAPEX avoided), improved profile reproducibility across campaigns. Estimate to validate.

And the fair question from the production manager

"Can a camera read a panel display reliably enough to trust the curve?" — the screen is fixed, the fields sit in known positions and fermentation temperatures move inside a narrow physical range, so it is an anchored task where the best models drop below 1.5% error [1]. Any frame whose value jumps outside what a tank can physically do in a few minutes is discarded rather than stored.

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