Horizon Europe projects analysis tool
Client
The problem
Every time Mosaic wants to draft a proposal for a Horizon Europe project, we need to find state-of-the-art technologies and projects in development in our area. Additionally, the task of finding candidates for the consortium is quite tedious. And even though all the information about projects, fundings and participants is publicly available, being able to make custom searches is not possible.
The solution
The purpose of the tool is to retain knowledge from a corpus of open documents from Horizon Europe projects.
It includes semantic search to answer questions about different aspects of the european projects.
Leveraging the growing popularity and technological advances in Generative Artificial Intelligence, the solution proposes using Large Language Models (LLMs) such as OpenAI’s ChatGPT, to transform natural language questions into customised queries for a data source.
Thanks to the logical capabilities of these models, complicated searches can be carried out automatically by being split up into more simple queries until all the necessary information is gathered.
-
- The tool allows searching information on individual projects using natural language.
- It includes filters to search for projects by metadata fields such as: total cost, start date, end date, duration, status, coordinator, etc.
- The LLM creates the appropriate query and selects the information to be returned as output from each project.
- Finally, it summarises the info into an answer in natural language.
- The chat history between User and LLM is stored and passed as context at each request.
- The conversation between LLMs is not stored.
Data
Through web scraping of data that can be found online at https://cordis.europa.eu/projects.
Results
We have saved up to 25% of initial research time for the preparation tasks.