Everything the cellar master sees, without stopping work

The problem

The finest knowledge in the winery walks out at the end of the shift.

Critical fermentation and aging know-how lives in the cellar master's head and never enters the system, unless someone stops to write it down.

• The cellar master spends the day between tanks and the barrel cellar with full hands: a hose, a tasting glass, a sampling pipette, a flashlight for the bunghole.
• They notice what no sensor sees: a barrel with a hint of volatile acidity, a tank that needs racking sooner, an SO2 top-up that should not wait until Monday.
• Writing it down means stopping, walking to a desk and opening a system. So it ends up in a pocket notebook, or in memory, or nowhere. By the end of the week, most of it is gone.
• That is how fermentation and aging know-how stays in one head, and how each campaign starts by relearning what the previous one already knew.

How it fits the IRIS system

Connect in voice mode — a wake word, the dictation, and the incident tied to its barrel.

Connect voice mode. Custom wake-word. The LLM structures the dictation as an incident linked to the active tank/ barrel/campaign. Reply by voice or push notification.

It only records what the cellar master chooses to say, when they choose to say it. That is what keeps the habit alive in the third week of harvest, and what turns a remark made in passing between two barrel rows into a record the next campaign can read. It is also the case that most changes how younger cellar staff learn, because the reasons behind decisions are finally written down.

See the full IRIS architecture →

Before and after

The notebook in the pocket versus the dictated record

Aspect	Today	With iLEAN Connect
An observation in the barrel cellar	Remembered, or not	Dictated on the spot
Which tank or barrel it refers to	Implied	Linked by name or number
Racking and SO2 decisions	In the cellar master's head	Recorded with their reason
Crossing with fermentation data	Impossible	Matched against the live curve
The cellar master's effort	Stop, walk, type	Speak without putting anything down
Conversations in the cellar	—	Not recorded

knowledge dies at shift end → enters the system in real time and gets matched against live fermentation data.

Impact estimate

Why we do not put a payback range on this case.

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 fixed payback range: the value is hard to put a number on upfront, and we would rather say so than invent one.
• What it protects is continuous improvement of fermentation and aging, which depends on the cellar master's knowledge existing in the system and not only in their memory.
• Every observation is matched in real time against live fermentation data, so patterns show up across tanks, barrels and campaigns.
• The estimate is built with the winemaker, starting from the racking and topping decisions that today depend on memory alone. Once a few months of records exist, the value becomes visible in decisions that are made faster and with fewer repeated mistakes.

hard to monetize upfront, but continuous improvement of fermentation and aging depends on this knowledge existing in the system. Estimate to validate with the winemaker.

And the fair question from the production manager

"It will not cope with cellar jargon and the echo of a vaulted barrel room." — recognition is tuned for industrial noise and for the winery's own words: tank numbers, barrel rows, racking, topping, lees. Structuring a dictated sentence into tank or barrel, observation and urgency is an anchored task, where the best models drop below 1.5% error [1], and what was understood is read back so it can be corrected with one word.

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