Healthcare & Life Sciences

With a provider certified for health data hosting, or inside your own facility where you have the infrastructure. An open-weights model operated in that perimeter covers most internal use cases, which avoids sending patient data to an external service. Worth being precise about what the certification covers: it applies to the hosting, not to what you do with the data. The purpose of the processing, its legal basis and patient information remain decisions for the organisation, and we document them with you rather than in your place.

No, and it is better to state that limit up front. Software intended to inform a diagnosis or a treatment is a medical device under European regulation 2017/745, which requires a manufacturer, a quality system and a notified body. We work upstream and alongside: data preparation and structuring, convenience tools that carry no diagnostic purpose, and support to a vendor who does hold the marking. If your need genuinely falls under medical device rules, we say so rather than working around it.

The timing has moved. Regulation 2026/1744 of 24 July 2026 pushed the obligations for high-risk systems to 2 December 2027, and to 2 August 2028 for AI embedded in an already regulated product, which is the case for software falling under the medical device regulation. An administrative convenience tool does not fall in that category; a tool that informs a clinical decision does. The sorting is done use case by use case, and it is usually the first piece of useful work.

Two deadlines of different kinds. The French Ségur programme is funding tied to referenced software versions: its second wave was launched by decree on 3 March 2026, and organisations had until 30 September 2026 to commit their project and secure the funding. The European Health Data Space is a regulation: in force since 26 March 2025, it applies in stages through to 2035, with health data access bodies designated in 2027 and secondary use opening in 2029. What prepares you for both is the same work: making your data exportable in a common format, with no manual rework.

Yes, on the data chain rather than on the science. Three subjects recur: connecting instruments to the laboratory information system so manual rework between devices disappears; making a processing chain replayable, from raw data to published result, because a result that cannot be reproduced cannot be defended at inspection; and documenting software validation during development instead of reconstructing it at the end. It is the same craft as in a care setting, applied where the evidence has to be kept far longer.

By working on free text, which is the real weak point: reports contain names, dates and contextual detail that no column-level masking catches. We pseudonymise that text, then measure the residual risk rather than declaring it nil, and log access per record. That measurement file is what your data protection officer and your ethics committee will examine: zero risk does not exist, a measured and documented risk can be owned.

Framing takes 2 to 6 weeks depending on scope and the level of compliance required. A first version tested by a small group of clinicians follows in 4 to 10 weeks, depending on the state of your protocols and document base, which is almost always the deciding factor. Rolling out across several departments, including training, runs to 3 to 6 months. These ranges are the ones we apply everywhere, and each comes with what makes it vary.

One project, documented: at Pearl Dental Paris, automatic identification of teeth, anomalies and implants on panoramic radiographs cut dental charting time by 95% and data entry errors by 65%. The project received the Silver Medal at the IT Nights 2022. We would rather present that specific case than a sector average: in a sector where we hold one reference, quoting generic percentages would help neither you nor us.