Allogeneic TCR-T — one donor, several doses, several patients and a single custody chain.

The problem

A tree-shaped custody chain that no system draws in full.

In an allogeneic TCR-T, the industrial model changes compared with autologous: there is no longer 1 patient ↔ 1 product, there is 1 donor → N doses → N patients in N hospitals. Every milestone has its own system, and nobody joins them:

1. Donor — donor bank system, with its consent, its typing and its regulatory traceability.
2. ATMP batch — manufacturer's LIMS, with the ex vivo editing, the cell bank and the N split doses.
3. Distribution to hospitals — several cryogenic dataloggers, each with its transport, its cold chain and its signature on arrival.
4. Infusion per recipient — hospital pharmacy system at each centre, with its patient, its responsible physician and its clinical event.

When one dose triggers an adverse event, the urgent question is: which sibling doses are live and where are they? Reconstructing that through calls and emails across three hospitals can take days. The classic system works almost always — and that almost-always, in allogeneic ATMP, is the one that calls the EMA at three in the morning.

How it fits the IRIS system

iLEAN Tracer does not add one more system — it seals the cracks between the ones you already have.

The allogeneic TCR-T problem is not a lack of data: the donor bank, the manufacturer's LIMS and each centre's hospital pharmacy have the information. The problem is that it lives in heterogeneous islands and that at the critical moment — the adverse event, the change of recipient, the sibling dose — the question crosses centres and nobody has the authority to answer it in minutes. Tracer is the filler that seals those cracks, without asking you to change the LIMS, the donor bank or the hospital system.

Edge cross-checks dose against patient in every infusion room. Connect reads the dataloggers and the LIMS of each centre. The agent maintains the donor tree and, if a branch deviates, raises a flag at second zero. The person signs at each centre.

The three iLEAN pieces applied to multi-recipient allogeneic TCR-T:

• Edge — terminal with vision (CNN) in every cleanroom and every infusion room. It reads the dose UDI-DI, cross-checks it with the patient in front of it and triggers a physical hold if it does not match. It works without network — if a hospital loses WiFi, Edge keeps reading and holding, and syncs with the tree when the network returns.
• Connect — captures the donor bank, the ATMP manufacturer's LIMS, the cryogenic transport dataloggers and the hospital pharmacies of each receiving centre. And it captures from the outside whatever arrives by email, messaging app or phone call (slot change, delay, reported adverse event) at second zero, without anyone forwarding anything.
• Agent — maintains the donor tree with all its doses and recipients, respecting data isolation between patients. If a branch deviates, it does not send an email at 10pm: it raises the flag at the centres holding sibling doses in stock or scheduled, and puts in front of each centre's physician the minimum information needed to decide. The one who signs is still the person.

See the full IRIS architecture →

Before and after

Manual allogeneic custody chain vs. donor tree with iLEAN Tracer

Aspect	Bank + LIMS + 3 hospital pharmacies	With iLEAN Tracer
View of the donor tree	Reconstructed after the fact, by hand	Live, in a single view, to the millisecond
Dose ↔ recipient identity	Verified at the centre's pharmacy	Verified also at the point of infusion, with a physical hold
Adverse-event alert to sibling doses	Calls + emails over days	Flag at second zero, minimum data per centre
Dossier for EMA / GMP Annex 1	Rebuilt by hand per dose and per recipient	Dossier per dose and per recipient, automatic and signed
Data isolation between patients	Assumed, with no technical enforcement	Configurable per ring, with cryptographic signature
Final decision to infuse	Physician, without the tree's context	Physician, with the tree and the siblings 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 receiving hospitals and your LIMS. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• ATMP plant with allogeneic TCR-T in production, 2-4 active receiving hospitals, sibling doses in simultaneous use at different centres.
• Tracer pilot in one cleanroom + two receiving hospitals (Edge at the bank and in the infusion room + Connect against LIMS, donor bank and hospital pharmacy + tree agent). First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on the average cost of a dose-patient identity near-miss and of the response time to a documented adverse event in recent batches.
• The hard lever is responding in minutes instead of days when a sibling dose triggers an event — a single episode covered in time pays for the pilot. Expected reduction in response times ≥ 30% once stabilised.

And the Quality Director's reasonable doubt

“What if the AI gets it wrong and mixes data between patients?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely maintains a tree and re-contextualises identifiers from one system to another, the best models brought error below 1.5%^[1]. And even so, what is critical is not decided alone: Tracer holds the infusion and the person signs. The three rings guarantee technical isolation between patients, and the system fits the EU AI Act and GMP Annex 1.

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