CAR-T against B-cell lymphoma — the cell is the patient and the window cannot be repeated.

The problem

Four clocks that have to add up, and no system carries them together.

An autologous CAR-T against B-cell lymphoma is not a drug batch — it is a single dose for a single person, and the raw material is that person's own cells. That uniqueness breaks the standard industrial model:

1. Apheresis — collection happens at the hospital, inside the hospital's system. The product's time zero starts there.
2. Cryogenic transport — the liquid nitrogen datalogger travels with the product, but its data usually reaches the ATMP manufacturer's site when it is already too late.
3. Ex vivo editing and cell bank — inside the ATMP manufacturer's system, under GMP Annex 1 and the viability window of that specific product.
4. Infusion — back to the hospital, to another system and another person — the responsible physician who signs.

Four systems, four clocks, no common brain crossing them against the product's viability window. The classic setup works almost always — and in autologous ATMP, the almost-always are exactly the ones that call the EMA and the quality team at three in the morning.

How it fits the IRIS system

iLEAN Tracer is not a fifth system — it is the filler between the four you already have.

The CAR-T problem is not a lack of data: the cryogenic datalogger, the manufacturer's LIMS, the hospital's ERP and the hospital pharmacy system do hold the information. The problem is that it lives on four islands and, at the critical moment — the viability clock — it does not reach the decision-maker in time. Tracer acts as the filler that closes those gaps, without asking you to change your LIMS, your ERP or the hospital's system.

Edge sees the product in the infusion room. Connect reads the cryogenic datalogger wherever it lives. The agent cross-checks the viability window and, if something does not add up, it holds. The person signs — never the other way round.

The three iLEAN pieces applied to CAR-T against B-cell lymphoma:

• Edge — a vision terminal (CNN) in the cell bank cleanroom and in the infusion room. It reads the product's UDI-DI and the patient identifier, cross-checks them, and triggers a physical hold if they do not match. It works with no network: if the hospital loses WiFi, Edge keeps reading and keeps holding, because what is critical cannot depend on connectivity.
• Connect — captures the cryogenic transport datalogger, the manufacturer's LIMS, the hospital's ERP and the hospital pharmacy system, whether they arrive through a modern API or through a spreadsheet exported at 10pm. And it captures from the outside whatever arrives by email, messaging app or phone call (a change in the surgical slot, a transport delay) at second zero, without anyone forwarding anything.
• Agent — cross-checks the four milestones against the product's viability window and against GMP Annex 1. If a deviation breaks the window, it does not send an email: it holds the infusion and alerts the responsible physician through whatever channel they use. The person validates and signs; the pump does not start on its own.

See the full IRIS architecture →

Before and after

Manual CAR-T custody chain vs. custody chain cross-checked with iLEAN Tracer

Aspect	Four systems + email	With iLEAN Tracer
Apheresis window clock	Reconstructed afterwards from 3 LIMS	Live, to the millisecond, on one screen
Patient ↔ product identity	Verified at the start and at the end	Verified at every milestone, including at the bedside
Cryogenic thermal deviation	Alert once the product is already on site	Alert before the infusion, with the viability curve
Dossier for EMA / GMP Annex 1	Rebuilt by hand, weeks of work	Per-patient dossier, automatic and signed
Operation with no network at the hospital	n/a	Edge keeps running on the panel's own power
Final decision to infuse	Physician, without the four clocks in front of them	Physician, with the four clocks and the window in front of them

Impact estimate

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

The block below is an estimate to be validated with the actual data of your cleanroom, your collecting hospitals and your LIMS. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• ATMP site with autologous CAR-T in production, 2-4 collecting hospitals and a single receiving centre, product viability window in the range of the critical hours.
• Tracer pilot in one cleanroom + one collecting hospital (Edge at the bank and in the infusion room + Connect against the LIMS and the cryogenic datalogger + viability agent). First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on the average cost of a documented near-miss in your recent batches and the weight of the EMA inspection dossier.
• The hard lever is a single batch rescued from a cryogenic near-miss — the raw material is the patient and collecting again may not be an option. Expected reduction of indexable near-misses ≥ 30% once the pilot stabilises.

And the responsible physician's fair objection

«What if the AI gets it wrong and lets through an infusion that breaks the window?» — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely recontextualises a data point from one system into another (reading a datalogger and comparing it against that product's viability curve), the best models brought the error below 1.5%^[1]. And even so, what is critical is never decided alone: Tracer holds the infusion and the responsible physician signs. The three security rings are there precisely for this, and they fit the EU AI Act and GMP Annex 1.

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