An innovative startup transforming the drug market access process through generative AI.
KEY FIGURES
Over 1000 reference documents from NICE (National Institute for Health and Care Excellence) of more than 400 pages
95% accuracy in responses
30 seconds response time
ABOUT THE CLIENT
Founded by Virginie Simon (MBA '24J), MxAI won the Grand Prize at the 48th edition of the INSEAD Venture Competition, along with €25,000 in funding. Based at Station F, the startup is revolutionizing the approach to drug market access by leveraging the potential of generative AI
TECHNICAL CHALLENGE
Managing and analyzing over 1000 pharmaceutical documents of 400 pages from NICE (National Institute for Health and Care Excellence) presented a major challenge. The complexity lay in the need to understand and extract relevant information from dense technical documentation while ensuring absolute accuracy, critical in the pharmaceutical field.
The complex medical terminology required a sophisticated approach, combining technical expertise and professional validation. The system needed to not only understand specialized language but also provide quick and accurate responses.
SOLUTION
AI Model
Use of Gemini for analysis and generation
Specialized processing of medical terminology
Data Management
MongoDB Atlas for vector storage
Optimized chunking for complex medical documents
Database consisting of 300 official NICE documents (National Institute for Health and Care Excellence)
Clinical guidance and therapeutic evaluation documents
RAG System
Retrieval-augmented generation architecture
Advanced contextual search
Specialized processing of medical documents
CONCRETE EXAMPLE
To illustrate the power of our system, let's take the case of a new drug in development. It's an innovative treatment for a rare neurological disease affecting children, characterized by epileptic seizures and progressive loss of motor skills and language. This particularly severe condition affects approximately 200 patients worldwide, with about a hundred new cases diagnosed each year.
The proposed treatment requires intrathecal administration over two days of hospitalization. Ongoing clinical studies (phase I/II) primarily evaluate the reduction in seizure frequency and monitor neurodevelopmental evolution through various standardized scales.
By analyzing its NICE document database, our AI identified three comparable treatments, each with its specificities:
Cannabidiol stands out for its simple oral administration and proven effectiveness on seizures, but requires close hepatic monitoring due to potential drug interactions.
Fenfluramine shows remarkable results in seizure reduction, with significant experience in its use. However, its profile requires regular cardiac monitoring and imposes certain dietary restrictions on patients.
Cerliponase has shown encouraging results in preserving motor skills and language. However, its intrathecal administration mode and high cost are constraints to consider.
This comprehensive analysis, generated in just 30 seconds, relies on NICE-validated data, allowing medical and regulatory teams to make informed decisions quickly.
METHODOLOGY
The project was completed in one month by a tight team consisting of an AI engineer and a developer, with expert validation from Virginie Simon. The iterative approach allowed continuous refinement of system accuracy through embedding benchmarks and expert validations.
RESULTS
The platform now achieves 95% accuracy in its analyses, enabling significant reduction in file processing time. The system efficiently processes 400-page documents, automatically extracting and analyzing relevant information.
INFRASTRUCTURE
The solution is deployed on a cloud infrastructure, ensuring scalability and performance. The chosen architecture allows smooth scaling while maintaining stable response times.
PERSPECTIVES
MAxAI continues to innovate in applying AI to the pharmaceutical sector. The success of this first phase paves the way for new developments, with the constant goal of improving and accelerating the drug market access process.