- Health
- IA
- Use cases
Using AI to analyze pharmaceutical patent data
The Customer
Pharmaceutical Group
Background and objectives
Our customer has embarked on a pioneering Generative AI initiative whose aim is to process large sets of unstructured data from biomedical literature.
The objective is substantial, as it aims to increase the potential for the discovery of new treatments for rare or genetically-induced diseases.
The data processed comes from public hosting platforms - the EPO (European Patent Office) and the WIPO (World International Property Organization) - where the number of patents can be counted in the tens of thousands.
Results
- The solution saves considerable time for data scientists working in the pharmaceutical group's R&D department.
- This is a major competitive advantage, as R&D teams benefit from significant added value in the search for new treatments.
Approach / System
Our AI solution enables you to consolidate data that is disparate in terms of type, source or reason for publication in the biomedical literature, into a structured whole that brings together the targeted information.
We have opted for a solution that focuses on certain areas of the literature, in order to emphasize the quality of the structured data provided to R&D teams rather than its exhaustiveness.
This structured data is a new source of knowledge for the algorithms used by the R&D department. It's a real lever for potential scientific discoveries.
Given the sheer volume of data - over 20 years of scientific literature - and the diversity of sources and typologies, the AI solution is not simply a matter of interacting with an LLM (Large Language Model). We use LLM interaction techniques based on several AI agents forming a team, where each is assigned a specific task and provided with tools to do so.
Added value Silamir Group
AI solutions can present a budgetary risk, as they generally rely on the use of paid APIs. At Silamir, we are well aware of this problem, and adopt a FinOps vision to keep the cost of our solutions in check.
For solution hosting, we recommend open source models to avoid dependence on paid APIs and keep costs under control.
