Objectives

To leverage knowledge from Earth observation and related sciences by pretraining on open-source EO documents, peer-reviewed papers, websites, and licensed data through specific agreements.

Develop and deploy an open-source LLM dedicated to Earth observation and related Earth sciences. Open-source a family of base and instructed LLMs for EO, as well as the training and benchmark datasets, specific to EO. All the source code of the development of EVEwill also be open-sourced.

Release an EO Virtual Expert, capable of Question-Answering, Document Summarization, and Schema Generation. The model is connected to a curated and updated RAG database of EO documents, which it will use to cite any empirical claims needed to answer a user query. A dedicated hallucination mitigation module will oversee all answers from the LLM to ensure factuality.

Equip users with open access to EO knowledge via an an API and a web interface.

Compliance is prioritized by ensuring proper licensing, anonymizing data to protect privacy, and maintaining transparency in AI development. Our project adheres to GDPR and open-source principles, with ongoing risk management and collaboration with legal experts.