Saana: a sovereign medical LLM, already used by hospitals
The problem
Medicine generates an enormous volume of content: clinical protocols, therapeutic guidelines, patient information, consultation reports. This content is written by and for healthcare professionals. It is precise and rigorous — and often incomprehensible to the vast majority of patients.
The result: a patient leaving a consultation with a chronic condition rarely has the resources to understand their illness, their treatment, or the day-to-day steps they need to follow. This gap is far from trivial. It has a direct impact on treatment adherence, the trust relationship with caregivers, and ultimately on clinical outcomes.
Saana's founding insight was simple: artificial intelligence can bridge this gap. But that insight immediately runs into a structural constraint of the sector — health data cannot be entrusted to just any general-purpose AI.
The major generalist AI platforms — however powerful they may be — are not designed for medical confidentiality. They store data in jurisdictions outside the control of care institutions. They offer no guarantee that the information exchanged will not be used to train their models. For a hospital or clinic, relying on these tools without precaution amounts to outsourcing the trust patients have placed in the institution — something no serious medical director can accept.
A different approach was needed.
The solution: Saana
Saana is a sovereign vertical LLM, specialized in medicine. It is not a general-purpose chatbot with a healthcare layer bolted on. It is a model designed for the medical domain, trained and operated under conditions that meet the specific requirements of that sector.
One of the first concrete use cases is a patient-facing application. It takes educational content produced by clinics — pedagogical resources that are often technical and dense — and transforms it into clear, accessible information adapted to the comprehension level of a non-medical person.
In practice: an institution already has a corpus of content validated by its medical teams. Saana does not reinvent that content — it makes it usable by the very people it is intended for.
What distinguishes Saana from a conventional AI project is that data hosting and control sit at the heart of the architecture, not as an afterthought. The infrastructure is sovereign. Data remains under the institution's control. The model leaks nothing to third-party systems.
And most importantly: Saana is already in production. Hospitals in Montpellier are using it today. This is not a prototype, not a proof-of-concept presented in a demo. It is a product that runs in a real environment, adopted by healthcare professionals.
Why sovereignty was non-negotiable
In many sectors, data sovereignty is a best practice. In healthcare, it is a legal and ethical requirement.
GDPR imposes strict rules on the processing of personal health data — classified as special-category sensitive data. Medical confidentiality adds an additional layer: a care institution bears responsibility for how that information is managed, including when it is entrusted to a technology provider.
But beyond the legal framework, there is a question of institutional trust. A hospital that adopts an AI solution does so in full view of its patients, its medical teams, and its governing bodies. If the question "where does our data go?" cannot receive a clear and documented answer, adoption is blocked — regardless of the product's quality.
Saana treated this question as a design priority, not a problem to be solved last. This is what made real deployment in hospital institutions possible.
The CTO role: architecture, product, and production deployment
Dibrilou Diagne is CTO of Saana. His role spans both the technical and product dimensions — two sides of the same challenge in a project of this nature.
On the technical side, this means designing an architecture that enables a specialized LLM to be trained and operated in a sovereign environment. These choices carry deep implications: selection of infrastructure, providers, security protocols, and data processing pipelines. These are decisions that commit the project for years to come.
On the product side, this means translating a real medical need into concrete features, navigating the constraints of medical teams, patient expectations, and the realities of deployment in hospital institutions — organizations with their own processes, their own information systems, and a very high reliability requirement.
This dual role — architect and product lead — is at the core of what Dibrilou brings. With eleven years of IT experience, a background in data (data hub and API for MGEN, health data platform at Air Liquide), and experience training technical teams at scale — including some fifty people on AI topics — he embodies a rare skill set: someone who understands both technical depth and the strategic product stakes.
Saana is proof that this combination delivers concrete, measurable results, in production.
What this project proves for your AI initiatives
There is a widespread belief in organizations exploring AI: the use cases are promising, but sector-specific constraints — regulatory, ethical, organizational — push production deployment far into the future or make it prohibitively complex.
Saana proves the opposite.
A specialized, sovereign LLM is not only technically feasible — it is being adopted by professionals in one of the most demanding sectors that exists. This is not a POC. This is not a laboratory demonstration. It is a product in production, in hospitals, today.
What Saana demonstrates:
- Specialization is an advantage: a model designed for a specific domain outperforms a generalist used without precaution in that same domain.
- Sovereignty is not an obstacle: addressed from the design phase, it becomes a differentiating asset and a driver of adoption.
- Production deployment is achievable: with the right architecture and the right product leadership, an AI project can reach real deployment, even in a constrained environment.
These principles are not unique to healthcare. They apply to any sector where data is sensitive, where compliance is demanding, and where end-user trust conditions adoption.
Let's talk about your project
Twenty supports leaders and teams who want to integrate AI into their organization without losing control of their data or their product.
The approach: the expertise of a large group, the closeness of a partner. You delegate technical complexity. You retain control over the decisions that matter.
If you are leading an AI project — in healthcare, in services, or in any sector where data is strategic — and you are looking for someone who has already done it under real conditions, let's talk.
WhatsApp: +33 6 34 42 50 56 Email: contact@twentyconsultancy.com