Syrup filling control with AI — one milliliter out of range is not an incident, it is an OOS deviation.

The problem

Three plant-floor realities that almost never reach the same person in time.

On a syrup filling line — solution or suspension, pediatric or adult — the guarantee of a correct dose rests on three pieces of data that in most plants each live on their own island:

1. The filler setpoint and curve — what volume the machine is calling for and how it is varying bottle by bottle. Sometimes in the PLC, sometimes only on an old local panel, sometimes in a spreadsheet kept by the formulation lead.
2. The real viscosity of the product in the hopper — temperature, agitation, API batch, sugar excipient batch. Almost never correlated in real time with the dose coming out.
3. The image of the bottle before the cap — real level, meniscus, bubbles, a drop hanging from the neck. The only thing the operator can see, and only on the bottles they have time to look at.

The quality manager knows everything adds up for 99% of the batch. It is that 1% that triggers the OOS investigation, the rework, and the doubt over the rest of the batch. Caught late it is a deviation; caught in line, it is nothing.

How it fits the IRIS system

iLEAN does not add a fourth system — it seals the cracks between the three you already have.

The syrup filling problem is not a lack of information: it is information living on islands that, at the critical moment (a change in syrup temperature, a new API batch, an agitator losing torque), does not reach the person who decides in time. iLEAN acts as the putty that fills those gaps, without asking you to change the dosing unit or the filler.

Edge sees the dose and the bottle neck before the cap. Connect reads the filler wherever it lives — modern PLC, legacy panel, forgotten spreadsheet. The agent cross-checks against the recipe and the GMP specification and, if something does not add up, holds the run. The person signs — never the other way round.

The three iLEAN pieces applied to syrup filling:

• Edge — a terminal with machine vision (CNN) over the filling carousel. It reads level, meniscus, bubbles and neck cleanliness in milliseconds and fires the actuator (bottle reject gate, stack light, selective stop) before the cap goes on. It works with no network. If the plant loses its Internet connection, Edge keeps reading and holding, because what is critical cannot depend on WiFi.
• Connect — captures the dosing unit whether the data comes from the PLC over OPC-UA, from an old terminal via screen OCR, or from the formulation lead's setpoint spreadsheet. And it also captures what arrives from outside (an email from the validation lead with a change to the viscosity range, a WhatsApp from the supplier about a deviation in the API batch) at second zero, without anyone having to forward anything.
• Agent — cross-checks the requested dose, the real viscosity, the image of the bottle, the MES recipe and the GMP specification. If there is a deviation, it does not send an email at 10 p.m.: it holds the run and alerts the quality manager on whatever channel they use. The person validates and signs; the line does not restart on its own.

See the full IRIS architecture →

Before and after

Filling controlled by checklist vs. filling cross-checked with iLEAN

Aspect	Manual control + sampling IPC	With iLEAN Edge + Connect + Agent
Dose verification	Periodic sample on a balance, the rest assumed	Every bottle verified in line, in milliseconds
Product viscosity	Measured at batch start, assumed stable	Correlated with the dose bottle by bottle
Dirty neck / drop	Detected if the operator happens to see it	CNN vision before the cap, selective reject gate
Deviation detection	At IPC or, worse, at release	Before sealing, with an automatic hold
Operation with no network	n/a	Edge keeps running on the cabinet's own power
File for an FDA/EMA auditor	Rebuilt from paper, days of work	Per-batch dossier, automatic, with images and traces

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

• Mid-sized pharma plant, one pediatric syrup filling line (solution + suspension), 40-60 bottles per minute, multi-SKU with format changeovers.
• Edge pilot on one line (camera over the carousel + actuator + integration with the dosing unit and the MES). First value expected within a few weeks: detection of dose deviation, dirty necks and bubbles above the manual standard.
• Reduction of OOS deviations caused by filling of ≥ 30% in the first pilot batch, depending on initial severity and the current spread of the dosing unit.
• Indicative payback between 4 and 9 months, depending on the frequency of documented OOS incidents in recent years and the average cost of rework/destruction in your plant.
• The hard lever is one single OOS deviation avoided: investigation, rework, risk of a scrapped batch, the auditor's attention. One major deviation pays for the pilot.

And the quality manager's reasonable doubt

"What if the AI gets it wrong and lets a badly filled bottle through?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely recontextualizes a piece of data from one system into another (reading the meniscus and comparing it with the setpoint), the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN holds the batch 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.

← Pharma blister control with AI · Suppository manufacturing control → · Nasal spray control with AI →