A spine patient leaves the consultation room with a diagnosis, a treatment plan, and a hundred questions they forgot to ask. Between that moment and their next appointment, they navigate alone — searching symptoms online, reading contradictory forum posts, unsure whether their pain is recovery or progression. Clinicians have a word for it: errance — the wandering from specialist to specialist, the silence between appointments that costs patients and health systems far too much.

Lamina exists to close that gap.

A Clinician-Led Partnership

Lamina is built by two teams with a shared conviction: most medical AI is built wrong. SpineDAO, a research collective of spine clinicians, scientists, and engineers, partnered with pleias, an open AI company building specialised language models around compliance, transparency, and European data sovereignty.

Lamina was designed to be different from day one: clinician-led, co-founded by orthopaedic clinical researcher Virginie Lafage, PhD; built on open, auditable models; and organised around clinical pathways rather than chat sessions.

VivaTech 2026: Small Models, Compliant Systems, Care Pathways

June 19, Paris. At the AI Factory France stage, Aleksandra Smilek hosted a conversation that cut through a week dominated by large-model benchmarks with a quieter, harder question: what does it actually take to deploy AI safely inside a clinical care pathway?

Virginie Lafage (Orthopaedic Clinical Researcher, co-founder of Lamina), Anastasia Stasenko (CEO, pleias), and Yannick Detrois (AI Scientist, pleias) gave three versions of the same answer. Small, open, and auditable models are not a compromise — they are a deliberate design choice that makes compliance tractable, data governance real, and clinical trust earnable.

For Lamina, that is not a theoretical position. It is a live architecture.

Why a Chatbot Is Not Enough

Most patient-facing medical AI is a chatbox with a disclaimer at the bottom. It can discuss anything. It has no concept of triage. It does not know when to reassure and when to escalate, and it forgets every conversation the moment it ends.

Lamina is built differently. Deterministic clinical logic governs onboarding, triage, and red-flag screening. Validated clinical instruments — the Oswestry Disability Index, STarT Back, and yellow- and red-flag detectors — run behind a conversational interface that feels natural but is, beneath the surface, structured. Context persists across appointments, and at the next visit the clinician receives a structured summary of what happened in the gap.

That is what pathways means in practice: not a workflow diagram, but a system that holds state across the silence between appointments — and returns something useful to the clinical team.

Inside the Architecture

Lamina is built with pleias on compact, open models trained entirely on documented open datasets — many produced on Jean-Zay, France's national supercomputer operated by GENCI and IDRIS. No patient data leaves controlled environments.

Both evaluation and generation draw on the Nemotron Personas series — statistically grounded synthetic persona datasets covering France, Belgium, and the United States, produced by pleias in collaboration with NVIDIA. For Lamina, the personas serve two purposes.

First, evaluation: testing model behaviour across the full demographic spectrum of patients the system will actually encounter — region by region and profile by profile — rather than a notional average user: the patient in Eupen whose home language is German, the Francophone patient in Liège, the Brussels professional who moves between both.

Second, generation: expanding a limited seed of real cases into demographically coherent synthetic populations — intake notes, clinical conversations, follow-ups — that reflect the genuine diversity a spine clinic sees.

"Better grounded data is what makes a small efficient regional-specific model viable in the first place." — pleias

The result is built for HIPAA and GDPR compliance, auditable end-to-end, and able to operate within the trust and cost constraints of a real clinical deployment.

What Comes Next

Lamina is heading into clinics and hospitals. The beta is being hardened ahead of first deployments — beginning precisely where the system can help patients most: in the gap between the consultation room and the next appointment.

If you are a clinician, healthcare institution, or partner organisation interested in what a properly built clinical AI pathway looks like in practice, we'd like to hear from you.

Follow SpineDAO