AI control of beer fermentation — the curve does not wait for daybreak, and the brewmaster cannot be on call all night.

The problem

Fermentation is a living process — and it lives while nobody is watching it.

The brewmaster knows their plant and their recipes. They know that an Ale starts fast in the first 24 hours, that a Lager demands low temperature and patience, that an IPA copes worse with a thermal spike on day three. But no brewmaster can be watching the tank at three in the morning, and fermentation problems keep no schedule:

1. Start-up failure — the yeast does not pitch properly (old batch, wort temperature off during transfer, poorly cleaned tank) and after 12 hours the gravity has not dropped as far as it should. If nobody sees it until morning, a day of tank time is already lost.
2. Thermal spike — the exothermic phase of primary fermentation releases heat; if the tank's cooling system partially fails and the operator does not see it until the morning reading, the ester and higher-alcohol profile changes and the batch comes out with a different sensory profile.
3. Stuck fermentation — the curve flattens out halfway to final gravity. There is room to rescue it (re-pitch, rouse gently, raise a degree), but only if it is caught in time.
4. Cross-contamination — a pH that climbs slowly instead of falling is the signature of bacterial contamination. If it is caught at 48 hours instead of 96, the problem can be isolated and cleaned before it reaches other tanks.

The classic system is the manual reading at the start and end of each shift: the operator climbs to the tank, draws a sample, reads gravity and temperature, notes it on a sheet, and the brewmaster reviews it when they can. It works most of the time. The problem is that a deviated batch only becomes visible once it is already done — and a badly fermented 100 hl tank is a hard cost that cannot be spread across the rest of the year.

How it fits the IRIS system

iLEAN does not replace the brewmaster — it gives them an AI that keeps watch while they sleep.

The valuable knowledge is in the brewmaster's head: what signature a well-started Ale has, how long a Lager takes on their specific recipe, when a deviation is noise and when it is signal. That knowledge is in no manual — it is the "veteran's eye", the biggest and worst-protected island of all. iLEAN acts as the putty that captures that pattern from the real batch history and puts it to work 24/7, without replacing the brewmaster: it only alerts them when there is a real deviation.

Edge reads gravity, temperature and time in every tank, continuously. Brain learns the signature of each recipe from the batch history. The agent alerts the brewmaster through whichever channel they use — before the deviation is beyond saving. The person decides and acts.

The iLEAN pieces applied to beer fermentation control:

• iLEAN Edge — a terminal in the fermentation hall that reads the existing sensors (inline gravity, Pt100 at several heights, pH, pressure, CO2 if you have them) and integrates them at the right frequency (every minute or finer). It works without a network: if the brewery loses WiFi in the small hours, Edge keeps reading and storing locally, because what is critical in a fermentation cannot depend on connectivity.
• Brain (Central + agents) — ingests your batch history (the real curve of each one with its final classification) and learns the characteristic signature of each of your recipes. It does not apply a textbook profile to you; it applies the profile your own brewery has demonstrated across its last few hundred batches. When a new batch drifts from that signature by a wide enough margin, it fires the alert.
• Connect (bidirectional alerts) — the alert reaches the brewmaster through whichever channel they use (phone, earpiece, tablet, WhatsApp), with the minimum useful information to decide in 30 seconds. The brewmaster replies through the same channel — confirms, dismisses or asks for more. Every decision feeds back into the agent.

The three safety rings guarantee that critical control over the fermentation (valves, cooling, transfers) stays in human hands: iLEAN proposes, the brewmaster signs.

See the full IRIS architecture →

Before and after

Manual readings per shift vs. continuous monitoring with iLEAN

Aspect	Manual reading + spreadsheet	With iLEAN Edge + Brain
Reading frequency	2-3 times a day per tank	Continuous (every minute or finer)
Hours covered	Working hours	24/7, including the small hours
Deviation detection	Whenever it shows up in a reading	The moment the curve separates from the recipe's signature
Knowledge used	Generic manual + the brewmaster's head	The real history of your batches (your own brewery's signature)
Alerting the brewmaster	The operator calls if they remember	Notification through the preferred channel with the key data
Batch file	A filed sheet of paper	Full curve + decisions + final classification

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your brewery. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• A brewery with 4-20 cylindroconical tanks (volumes between 10 and 200 hl), a varied recipe book (Ale + Lager + IPA + seasonals), an experienced brewmaster and a small team.
• Edge pilot on 2-4 tanks (the most sensitive ones or the largest by volume) + ingestion of the batch history to train each recipe's signature. First value expected within a few weeks — the alert on the first overnight deviation already shows in month one.
• Expected reduction of deviated batches of ≥30% against the baseline of the last few years (often more, but that is the defensible floor). Less brewmaster time spent on routine review, freeing them up for new recipes and development.
• Indicative payback between 4 and 9 months, depending on how often batches deviate and the average volume per batch. The hard lever is a single batch saved: a badly fermented 50 hl tank pays for a good part of the pilot.

The brewmaster's argument — the AI does not decide; you decide better

"Is the AI going to take my judgment away?" — quite the opposite. The brewmaster brings the judgment; the AI spares them the noise. Hallucination is a problem of free generation, not of anchored tasks. Comparing a real curve against a learned signature is exactly an anchored task — in this kind of task, the best models brought error below 1.5% ^[1]. And even so, what is critical (correcting temperature, extending fermentation, transferring) is signed by the brewmaster. iLEAN does not open valves; iLEAN alerts. Assist and simplify, not replace.

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