AI mRNA vaccine control — a batch is not signed off on hunches.

The problem

Three critical processes that almost never talk to each other within the same batch.

A signable mRNA vaccine is the clean cross-check of three processes that, in most plants, each live in a different system:

1. mRNA synthesis — T7 polymerase, capping, polyadenylation, purification by chromatography. The batch record lives in the vertical LIMS/MES or in the bioreactor's forms.
2. LNP encapsulation — microfluidic mixing of the mRNA with PEG-lipid, ionizable lipid, cholesterol and helper. Critical parameters (flow rates, ratio, temperature) in the skid's SCADA; size/PDI/encapsulation efficiency in the QC lab's DLS.
3. Annex 1 environment — particles, microbiology, filter integrity, operator training. A documented CCS that has to add up for every single vial.

The QC director knows it, but cannot be in three places at once. The operator has three different screens and a paper batch record signed across four shifts. The classic system works 99% of the time. That 1% is the OOS that reaches the FDA inspector — and in a vaccine that is not an incident, it is a campaign put on pause.

How it fits the IRIS system

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

The mRNA vaccine problem is not a lack of information: it is information living on islands — the LIMS, the LNP skid's SCADA, the environmental monitor, the QC LIMS — that at the critical moment (formulation, filling, release) does not reach the person who signs already cross-checked. iLEAN acts as the putty that fills those gaps, without asking you to change your validated MES, your LNP skid or your EBR system.

Edge sees the DLS and the filling line. Connect reads parameters from the LNP skid, the bioreactor and the environmental monitor. The agent cross-checks against the GMP recipe and Annex 1 — and if something does not add up, it holds the batch. The person signs — never the other way round.

The three iLEAN pieces applied to mRNA vaccine control:

• Edge — a terminal with machine vision (CNN) on the filling line and at the DLS bench. It reads nanoparticle size and PDI in real time, inspects the vial (volume, seal, stopper, label) and fires the actuator (rejector, stack light) if it does not match the recipe. It works without a network.
• Connect — captures the LNP skid parameters whether the SCADA is modern (OPC UA, MQTT) or an old industrial PC from the manufacturer. It captures the MES batch record and the Annex 1 environmental monitor. And it captures whatever arrives from outside — the ionizable lipid supplier's certificate, an alert from the transport data logger — at second zero, without anyone forwarding anything.
• Agent — cross-checks synthesis, encapsulation, environment and the GMP recipe. If there is a deviation, it does not wait for final QC: it holds the batch and prepares the deviation file with a cross-checked root cause. The QC manager validates and signs; the campaign does not restart on its own.

See the full IRIS architecture →

Before and after

Manual release of the mRNA batch vs. cross-checked release with iLEAN

Aspect	Manual batch record + offline QC	With iLEAN Edge + Connect + Agent
LNP skid parameters	SCADA on an island, daily CSV export	Second zero, tied to every vial
LNP size/PDI	Offline DLS, result in hours	Edge at the bench, result in line
Annex 1 environment	Separate vertical system, weekly review	Cross-checked per batch, immediate alert
OOS detection	Final QC after the whole fill	Before filling, batch held
Deviation file	Rebuild by hand, weeks	Cross-checked root cause, ready for CAPA
Operation without a network	n/a	Edge keeps running on the panel's own power

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.

• GMP mRNA vaccine plant with one or two LNP skids, a grade A aseptic filling line in RABS or isolator, a vertical Annex 1 environmental monitor, partial EBR.
• Connect + Agent pilot on the LNP skid (without touching the validated recipe) + Edge at the DLS bench. First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on the frequency of LNP-driven OOS in historical batches and the average cost of a lost batch in your market.
• The hard lever is a single batch rescued plus the QC time saved on every routine release. One campaign batch pays for the pilot several times over.

And the pharma CAIO's reasonable doubt

“What if the AI hallucinates about a DLS or SCADA reading and releases a batch it should not?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI recontextualizes a piece of data from one system to another (reading the DLS and cross-checking it with the GMP recipe), the best models brought error below 1.5% ^[1]. And even so, in pharma what is critical is never decided alone: iLEAN holds the batch and the person signs. The three safety rings exist precisely for this — ring 1, OT, is sacred; ring 2 validates; ring 3 proposes.

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