ISO 13485 with AI — the device file, alive and traceable from the part to post-market.

The problem

Four files that always have to add up — and almost never live in the same place.

The quality system of a medical device manufacturer has to keep four files alive and consistent — and most plants have them scattered across as many systems as they have departments:

1. DHF (Design History File) — the design history. It usually sits in a PLM/eQMS, but the real changes (scale-up to production, supplier changes) end up in another system or in an R&D spreadsheet.
2. DMR (Device Master Record) — the master recipe of the product in production. It lives close to the ERP/MES, but the approved deviations that modify it are not always updated at the pace of the plant.
3. UDI — the unique identifier of every unit, traceable to its batch and to its DMR and DHF. It is printed or marked on the line, but bidirectional traceability (from the unit to the batch, from the batch to its DHF) requires cross-checking three systems.
4. Post-market monitoring (PMS) — the complaints, incidents and reports that arrive through heterogeneous channels (distributor email, hospital call, web form). They almost never reach the system that cross-checks them against the batch and the DHF in time, and reverse traceability stays incomplete.

When the notified body audits, or when a post-market event forces a field corrective action, the cross-check of all four has to be rebuilt by hand. The honest sentence from the quality manager: "my DHF is frozen at the launch version, my DMR changes with every deviation, my UDI is run by production, and complaints are handled by the RA team — and all four have to add up in the dossier".

How it fits the IRIS system

iLEAN does not replace your eQMS — it seals the crack between the four files and reality.

A medical device quality system does not fail at documentation — it fails at the synchronization between what happens on the floor and the live files. iLEAN acts as the filler that closes that gap. It does not ask you to change your eQMS, your PLM, your ERP or your complaint system. It sits on top and guarantees that evidence arrives complete, linked to the right UDI/batch/DHF, and signed.

Edge reads the UDI and the integrity of the pack. Tracer captures every deviation, every complaint, every supplier certificate. Writer keeps DHF, DMR and PSUR alive. The person — from RA — signs.

The iLEAN pieces applied to ISO 13485 with UDI and MDR:

• iLEAN Edge — terminals with computer vision (CNN) over the medical device line: OCR of the printed or marked UDI, integrity of the sterile pack seal, labelling control, batch number and expiry date reading. If anything does not match what is expected, it holds the unit in milliseconds before release. It works with no network: medical device safety cannot depend on connectivity.
• iLEAN Tracer (Connect + integrations) — captures everything the eQMS cannot see on its own: every deviation an operator opens, every change request from the R&D team, every supplier certificate that arrives by email, every complaint that comes in through the web form or from the distributor by WhatsApp. It links each one to the batch, to the UDI and to the DHF/DMR element it touches.
• iLEAN Writer + RA Agent — Writer builds living dossiers for DHF, DMR, PSUR and MDR/QMSR. The agent cross-checks consistency across the four files and raises a flag when an approved deviation has not been reflected in the current DMR, when a PSUR needs updating because of the frequency of a complaint, or when a UDI registered in EUDAMED/GUDID does not match the batch produced. The RA person decides and signs — the line does not go to the field on its own.

See the full IRIS architecture →

Before and after

eQMS + regulatory sprints vs. living files with iLEAN

Aspect	Classic eQMS + sprints	With iLEAN Edge + Tracer + Writer
DHF between launch and annual review	Frozen — real changes never arrive	Alive — every change request updates the DHF
DMR vs. plant reality	Out of date whenever deviations are approved	Synchronized in real time with MES/ERP
UDI: bidirectional traceability	Manual — three systems cross-checked by hand	Edge reads + Tracer links + automatic dossier
Post-market monitoring	Scattered channels, late arrival	Unified capture at second zero
PSUR generation	3-5 weeks rebuilding evidence	Pre-built dossier — review and sign
Notified body audit	Pre-auditor sprint by the quality team	File ready every day of the year

Impact estimate

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

The block below is an estimate to be validated with the data from your plant and your regulatory portfolio. We lay it out so the committee has an order of magnitude; we refine it during the diagnostic.

• Class IIa/IIb medical device manufacturer with ISO 13485 + MDR + UDI, 1-3 production lines, an RA/QA team of 4-10 people.
• First value expected within a few weeks: Edge pilot on one line with UDI reading + pack integrity + unified capture of post-market complaints for one product family.
• Indicative payback between 4 and 9 months, depending on the monthly volume of deviations, change requests and complaints, and on the cost of regulatory sprints (PSUR, MDR Technical File update, notified body audits).
• Expected reduction in time spent on administrative evidence work: ≥30% (estimate to be validated). The hard lever is not the hours saved — it is the regulatory risk avoided: one major notified body finding or one FDA Form 483 costs far more than the entire pilot.

And the RA lead's reasonable objection

"What if the AI gets it wrong when linking a complaint to a batch or reading a UDI?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where AI merely recontextualizes a data point (reading a UDI and comparing it with the expected batch, normalizing a complaint that arrived by email), the best models brought error below 1.5% ^[1]. And even so, nothing critical is decided alone: iLEAN holds the unit or proposes the cross-check, and the RA person signs. The European AI Act and medical device regulation agree in requiring human oversight for high-risk AI — iLEAN's three rings are designed precisely to support that requirement without giving up the power.

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