Skip to content

EVE — Earth Virtual Expert

EVE Avatar Earth Virtual Expert (EVE) An open science initiative funded and supported by the European Space Agency's Φ-lab, developed by Pi School in collaboration with Imperative Space and Mistral AI, aimed at advancing the use of Large Language Models (LLMs) for the Earth Observation (EO) and Earth Science (ES) communities.

Getting started

In this documentation, you will find all the necessary information to understand, use, and contribute to the EVE project. The EVE project encompasses several components that have been open-sourced for the benefit of the EO community. In this guide we provide the documentation for the components we built and used as part of the EVE project:

  • Evalkit: A comprehensive evaluation framework for assessing LLMs on EO-specific tasks and General Domain benchmarks.
  • Data Scraping: Framework used to scrape relevant EO documents from trusted sources.
  • Data Processing: Comprehensive pipeline for cleaning, filtering, chunking and embedding EO documents.
  • Chat-Backend: Codebase that powers the backend of the EVE chat platform.
  • Chat-Frontend: Codebase that powers the frontend of the EVE chat platform.

All these components represent the building blocks of the EVE project and are designed to be reusable and extensible for future research and development in the field of LLMs for Earth Observation. You can navigate through the different sections of this documentation using the sidebar on the left. Each section provides detailed information, examples, and instructions to help you get started with EVE.

Other than these components, the EVE project also includes several repositories of experiments containing hands-on notebooks that have been included in the components above. These experiments repositories are not part of the main documentation since they do not contain production-ready code, however they are a useful entry point for those who want a hands-on experience exploring the EVE project. Here below the list of experiments repositories:

  • EVE-RAG: A collection of notebooks demonstrating how to build Retrieval-Augmented Generation (RAG) systems and how to benchmark it.
  • EVE-Hallucination: repository containing experiments and analysis on the hallucination pipeline implemented in EVE

Project Overview

EVE is a domain-specialized AI system designed to support scientists, analysts, and decision-makers through natural-language interaction.

It provides expert assistance across tasks such as:

  • EO and ES knowledge search and summarisation
  • Scientific Q&A and policy brief generation
  • Document understanding and retrieval
  • Technical and educational content creation
  • Integration of private EO document collections via RAG

The project unites data curation, model development, evaluation, and deployment to deliver open, reusable tools for Europe’s EO ecosystem.

Chat Platform

EVE features an interactive chat platform that allows users to explore and query Earth Observation and Earth Science knowledge through natural dialogue.
The platform is currently under closed access — you can request an early access to the beta version via this registration link here.
It is targeted for public release in early 2026.

Model Development

EVE-Instruct is trained on domain-curated corpora by fine-tuning Mistral Small 3.2 24B, through continued pretraining and instruction fine-tuning.

  • The model is optimized for factuality, reasoning, and integration with retrieval-augmented generation.
  • Training was carried out using EuroHPC Marenostrum, cloud environments, and Mistral infrastructure.
  • Evaluation targets both general domain and EO-specific benchmarks.

Licence

All EVE releases follow ESA and EU open-science principles, promoting transparency, reproducibility, and European AI sovereignty.
The project’s outputs—models, datasets, pipelines, and documentation—are shared under permissive open-source licences (Apache 2.0 / CC-BY).

  • 🌐 Project Website — eve.philab.esa.int
    Central hub for project updates, communication materials, and registration for early access to the EVE chat platform.

  • 🤗 Models & Data — huggingface.co/eve-esa
    Repository for domain-specific model checkpoints, datasets, and evaluation benchmarks.

  • 💻 Codebase — github.com/eve-esa
    Open-source tools, pipelines, and scripts developed within the EVE initiative.

Contributors

European Space Agency (Φ-lab) Pi School · Imperative Space · Mistral AI ·