Client work

CNGE helps doctors find medical information during consultations

For EBiM, Fasfox designed and integrated a Mistral AI assistant that helps doctors search sources validated by the College.

The Collège national des généralistes enseignants (CNGE, French National College of Teaching General Practitioners) trains general practitioners and contributes to the teaching of their discipline. As part of the EBiM project, it wanted to give its members quick access to validated medical information through an assistant using artificial intelligence.

Results

Challenges and objectives

Context

With EBiM, the CNGE aimed to provide its member GPs with a reliable, secure and high-performing generative AI tool to assist them in their practice. The goal was to create a genuine AI medical conversational agent that summarizes information and presents it in a clinically relevant way during the consultation. The chosen solution, Mistral AI’s Le Chat, was to rely exclusively on data sources validated by experts, in order to ensure the relevance and safety of the information provided, which is imperative in healthcare.

Key objectives

The project had three objectives:

  • Provide reliable AI: the tool must ensure accurate medical answers, based exclusively on data sources validated by experts, using a large language model (LLM) from Mistral AI.
  • Support doctors in delivering efficient, high-quality care: make information quickly accessible to improve decision-making during consultations.
  • Make the most of CNGE resources: build on the organization’s knowledge by feeding the solution with the CNGE’s validated document libraries and websites.

Solution and use case

The project centers on setting up the Le Chat environment and creating the “Quick search during consultation” agent.

This agent is the cornerstone of the project, designed to be used directly in front of a patient. Its effectiveness relies on the speed of Mistral AI’s technology and the relevance of the sources validated by the CNGE.

FeatureSpecification
Context of useQuick search during a consultation.
Response timeUnder 5 seconds
Expected answerAccurate and referenced, quickly providing essential information.
Information prioritiesDiagnostic hypotheses, differential diagnoses, prevalent and serious diagnoses, treatments, dosage, contraindications and drug interactions.
Level of expertiseMeet the needs of an expert primary care clinician.
Data sourcesValidated and reliable sources only; integration of the document libraries and websites provided by the CNGE.
Results

Technical architecture and method

The technical approach focuses on integrating validated data sources and configuring the conversational agent to ensure reliable AI answers.

  • Platform: Mistral AI’s Le Chat provides the conversational infrastructure.
  • Data sources: integration of the CNGE’s document and web resources, with a strict requirement for validation and reliability for all sources.
  • Data scraping for the AI: to integrate the CNGE’s document libraries and websites, ingestion scripts were developed. These scripts convert heterogeneous formats (HTML, etc.) into a data structure optimized for the Mistral AI engine, forming the knowledge base that feeds the conversational agent.
  • Dynamic automation mechanism: a system of data pipelines was developed to feed the validated sources provided by the CNGE continuously. These pipelines regularly and securely scrape all the resources (validated websites) to keep the agent’s knowledge base up to date with the latest data published by the CNGE.
  • Deployment: configuration of the environment and creation of the “Quick search during consultation” agent, prioritizing speed of execution and accuracy.

Results and benefits of the project

The medical conversational assistant brings concrete benefits to member GPs:

  1. Reliable information: the assistant ensures that answers are based exclusively on the CNGE’s validated knowledge, reducing the risk of errors linked to unverified sources.
  2. Time saved during consultations: with a required response time of under 5 seconds for quick searches, the tool lets doctors access critical information (such as dosage or interactions) without disrupting the consultation.
  3. Improved quality of care: by providing quick, targeted access to expert knowledge (diagnostic hypotheses, treatments), the platform supports clinical decision-making.
  4. Building on CNGE knowledge: integrating the organization’s document resources into the platform preserves and enhances the knowledge developed by the College for the benefit of its members.