Controlling the aging of Gorgonzola DOP in natural caves — humidity, CO2 and needles, without touching the cave.

The problem

A cave cannot be reprogrammed: if the batch goes wrong, you find out fifty days late.

Gorgonzola DOP is a slow cheese. The curd carries the Penicillium roqueforti that will produce the veining; molding is done in the characteristic molds, and two or three weeks later the wheels are pierced with long needles so that air can get in. Then into the cave: 50 to 150 days depending on whether it is dolce or piccante. And the result only shows up at the end.

In most Gorgonzola dairies that lives in three different places:

1. The piercing room — needle pattern, count, depth. Often verified by manual sampling.
2. The natural cave — humidity, temperature, CO2. Sensors that may be old or recent, data that almost never leaves the room.
3. The aging log — turning and brushing frequency per shelf, in a notebook or in a vertical MES.

The aging manager carries it all in his head. But when a cave holds five thousand wheels and three months of load, a silent deviation on one shelf can eat a batch without anyone seeing it. The damage is discovered when the wheel is cut — and by then it is too late.

How it fits the IRIS system

iLEAN does not touch the cave — it watches what goes into the cave and logs the environment.

The problem with aging is not a lack of information: it is information that lives on islands and only gets cross-referenced when the wheel is cut. iLEAN acts as the putty that fills those gaps, without asking you to change the piercing room, the cave or the aging system. The natural cave is left untouched — that is part of the DOP promise.

Edge sees the piercing inline. Connect captures the cave sensors and the turning logs. The agent cross-references them with the cahier des charges and warns the aging manager. The person signs — never the other way round.

The iLEAN pieces applied to Gorgonzola DOP aging in natural caves:

• Edge — a camera with AI vision (CNN) over the piercing room. It verifies the needle pattern (count, depth, distribution) on every wheel, in milliseconds. If a wheel comes out with an off-pattern piercing, it fires an actuator (stoplight signal, rejecter) and raises an alert. It works with no network.
• Connect — captures the existing humidity, temperature and CO2 sensors in each cave (through integration or a photo of the log), and the turning and brushing notebooks by photo or by the operator's voice. It also captures what comes in from outside (a rule change from the Consorzio Gorgonzola) at second zero.
• Agent — cross-references the piercing, the cave environment and the per-batch aging calendar against the cahier des charges. If a shelf has been out of humidity range for several hours, if a turning round has been skipped, if a piercing pattern has drifted — it does not send an email at 10 p.m.: it tells the aging manager right then, with the context.

See the full IRIS architecture →

Before and after

Manual aging vs. aging assisted by iLEAN Edge

Aspect	Notebook + sampling + the eye	With iLEAN Edge + Connect + Agent
Verifying the needle piercing	Manual sampling of a few wheels	Vision on every wheel, pattern verified
Cave environment (humidity/temp./CO2)	A sensor with a panel in the room	Integrated into the system, alerts per shelf
Turnings and brushings	Paper notebook, easy to forget	Captured by Connect (photo / voice)
Detecting an aging deviation	When the wheel is cut, 50-150 days later	Right away, to the aging manager
Operation with no network in the cave	n/a	Edge keeps watching on the power it has
Per-batch dossier for the DOP	Rebuilt by hand, weeks of work	Ready, with a piercing photo and environmental curves

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 dairy. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Mid-sized Gorgonzola DOP dairy, ~300-700 wheels/day, 2-4 active caves running dolce and piccante.
• Edge pilot over the piercing room plus sensor capture from one cave, with MES integration and a mapping of the Consorzio's cahier des charges. First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on how often batches are downgraded by aging deviations and on the average cost per cave-batch.
• Expected reduction in batches downgraded by aging or piercing deviations of ≥ 30%, with a per-batch dossier for the Consorzio.
• The hard lever is a single cave-batch not downgraded: at what a Gorgonzola cave is worth, one alone pays for the pilot.

And the aging manager's reasonable doubt

“What if the AI gets it wrong and lets a badly pierced wheel through?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely recontextualizes a sensory data point (verifying a needle pattern, comparing humidity against a range), the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN flags the doubtful wheel and the person signs. The three safety rings exist precisely for this.

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