It is estimated that medication errors are responsible for more than 140,000 hospitalizations and 13,000 deaths in France. Generative AI is a very promising tool for securing prescriptions while freeing up time for health professionals. Vidal, a leader in medical information in France, wants to develop a tool to provide accurate and reliable answers to PDS in record time thanks to GenAI, in order to optimize their medical time and ensure better patient care.
Context
Vidal, a reference platform for medical information on health products for more than 100 years, must modernize and digitize its services in order to maintain its leading position and remain competitive. They want to enter the functional area, that is to say to provide the answer to a specific question directly, in record time.
Generative AI is the most effective way to meet this challenge, and allow health professionals to focus on their care time. It is essential to assess the opportunities for using these tools. Vidal therefore needs to quickly demonstrate the value of investing in these new technologies.
Challenges
- Deliver a proof of concept in eight weeks on an AI chatbot based on the Vidal database.
- Save doctors time researching medical questions compared to going to the website.
- Gain the trust of health professionals in the answers provided by the chatbot.
- Train Vidal teams to maintain and improve the chatbot.
Our approach
A team of 4 Theodo experts supported the Vidal teams to carry out the mission (1 PM health expert, 3 LLM expert to develop the RAG).
What made the difference:
Value for client
The PDS gets a response 4 times faster, based on +30,000 monographs, which guarantees the reliability and richness of the information.
.svg)
Right-First-Time
To assess the answers, we developed a dataset of 100 questions with doctors, which will be tested throughout the project to validate the reliability of the model.
.svg)
Learning Organization
The composition of a hybrid team (product owner specialized in health, data expert - GenAI, web expert) and the joint work with doctors allowed us to continuously validate the model and to be agile in prioritizing the features to be developed.
%201.png)
“The support offered gave us all the support we needed. We knew that we had to get started with GenAI technologies, and we decided to get support from competent people. In fact, I am going to strengthen the team on Vidal's side and put in more developers so that we can continue to learn!”
