Skip to content
Raisona Contact
A cloud of thousands of small pages arranged by subject area, from which four cited pages emerge, each connected by a thread.

GRADeS e-santé Occitanie · Health

The documentation answers

From the trainer’s description of the course, the assistant we designed at Raisona searches the entire document collection. It then drafts an outline with live links to the relevant pages.

GRADeS e-santé Occitanie

The regional support group for e-health development in Occitanie, in the south of France, is known by its French acronym, GRADeS. A public interest group, it implements the regional e-health strategy on behalf of the Occitanie Regional Health Agency. Its 128 members range from the university hospitals of Toulouse, Montpellier and Nîmes to health and social care facilities and the regional unions of healthcare professionals.

The group provides health and social care professionals with digital services, from secure messaging to remote expert advice, and trains them to use those services. With the universities of Toulouse and Montpellier, it also helps train current and future professionals in digital health. We designed the assistant for the group’s trainers.

Adapting teaching as the field changes

In digital health, one change follows another: a standard moves to a new version, a service goes live, a security rule is tightened.

Even before entering the classroom, the trainer has to find, among guides, slide decks, reference texts and videos, what suits their audience and the planned length of the session. We designed the assistant for that preparation stage. Every document added to the collection feeds into its answers.

Core servicesCybersecurityTelehealthStandardstodayoutline
Subject area by subject area, the collection grows. Each outline is built from the documents it holds at the time of the request.

We focus on making it easy for the trainer to check that the content is valid.

Finding the source of an answer

Before searching, the assistant asks the trainer what it does not yet know: the audience, the level, the length, the format. It then searches the collection, by meaning and by keyword, and drafts the outline.

Each item in the outline points to its document and page, which open in a single click. The trainer can then read, keep or discard, and leave with their own selection.

Course outlineSource: guide, p. 12the documentp. 12the passage theoutline usedthe page
From a line in the outline to the passage in the document: the path of a citation.

Keeping a record of every search

Each search is recorded under a pseudonym, with the need expressed, the intended audience, the documents suggested and those selected. Search by search, a memory builds up: how a course is prepared, what the trainer keeps and what they leave out.

We are building this database from the outset so that future models can also learn from the choices trainers make.

NeedAudienceSuggestedSelectedlearning,later NeedAudienceSuggestedSelectedlearning,later
Each thread links a need to an audience and to a suggested document, then to a selected document when the trainer keeps it.

Keeping control of your data

The assistant is installed at the public interest group’s hosting provider, together with the documents and the search memory. The group keeps the code and the database, and only the request and the relevant excerpts are sent to the large language model (LLM).