Same reactor, now it talks

The problem

The curve that explains a batch's efficacy scrolls off the screen.

The synthesis reactor or formulation tank is 15 to 20 years old, with a proprietary panel that isn't networked. Replacing it means months of downtime and a huge investment. That screen shows the temperature curve, pH and agitation speed every few seconds: exactly what would correlate reaction conditions with the bioinput's final efficacy. Today nobody logs it.

• The fermenter or synthesis reactor is 15 to 20 years old and runs on a proprietary panel that was never networked. Replacing it means months of downtime and a very large investment.
• That screen shows temperature, pH and agitation speed every few seconds — exactly what determines whether a Bacillus or Trichoderma culture reaches the CFU count and viability the product promises.
• Nobody logs it. At best, an operator notes a reading per shift on a sheet, which says nothing about a temperature excursion in the middle of the night. A culture can recover its curve and still lose viability, and the sheet will show nothing unusual.
• So when a bioassay comes back low, the question everyone asks — what happened in the fermenter during that batch? — has no answer. The batch is reworked or discarded, and the same profile can repeat next month without anyone recognizing it.

How it fits the IRIS system

Connect in photo mode on the panel — the fermenter keeps its original control.

iLEAN Connect places a fixed industrial camera pointed at the panel. OCR with a grounded LLM interprets each frame and builds the historical curve batch by batch, without touching the original control.

Nothing is wired into the fermenter and nothing is installed on its controller: a camera looks at the screen the operator already looks at. For a vessel whose control nobody dares to touch, that is the difference between a project that gets approved and one that never starts. And it costs a fraction of the replacement that keeps being postponed.

See the full IRIS architecture →

Before and after

The fermenter panel unread versus the fermenter panel digitized

Aspect	Today	With iLEAN Connect
Temperature, pH and agitation	Visible for seconds, then gone	A continuous curve per batch
Night-time excursion	Nobody sees it	Stored with its time and duration
Curve versus bioassay result	Zero correlation	Crossed batch by batch
Root cause of a low-CFU batch	Guesswork in a meeting	Minutes
The fermenter's controller	—	Untouched, no rewiring
Vessel replacement	On the investment plan	Deferrable indefinitely

Before: zero correlation between the curve and bioassay results. After: when a batch is out of spec, root-cause analysis takes minutes. Estimated payback 5 to 10 months, and reactor replacement can be deferred indefinitely.

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.
• Batch-by-batch reaction curve traceability without replacing the fermenter, whose replacement can be deferred indefinitely.
• When a batch is out of spec, root-cause analysis goes from impossible to a matter of minutes, because the curve is there to be crossed with the bioassay.
• And, over time, the history shows which fermentation profile produces the best viability — knowledge that today lives only in the most experienced operator's head and leaves the plant with them.

Estimated payback 5-10 months · batch-by-batch reaction curve traceability without replacing the reactor.

And the fair question from the production manager

“Can I trust a camera reading a screen that old?” — the task is anchored: same panel, values always in the same position and physically bounded ranges, and there the best models drop below 1.5% error [1]. Any frame whose reading falls outside the possible range — a pH that cannot exist, a jump no culture produces — is discarded and flagged, not stored as data. The curve is built only from readings that make physical sense, and gaps are shown as gaps rather than filled in.

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