🛰️ ESA-SCEVA

ESA-SCEVA is the official Hugging Face organisation of the SatCom Expert Virtual Assistant (SCEVA) project, developed within the European Space Agency (ESA) ARTES programme.

The organisation hosts all open models, datasets, and tools produced under the SatcomLLM activity, titled
“Assessment of Open-Source Large Language Models within the SatCom Sector (ARTES FP 1A.128)”.

🌍 Mission

To develop, evaluate, and release open Large Language Models (LLMs) and datasets specialised for Satellite Communications (SatCom).
SCEVA aims to support engineers, researchers, and mission planners with advanced natural language capabilities for:

• Link budget analysis
• System and payload design
• 5G/6G non-terrestrial network integration
• Mission operations and anomaly detection
• Regulatory and cybersecurity analysis

The goal is to strengthen European digital sovereignty in AI for space through transparent, reproducible, and open research.

🧠 Key Deliverables

Models

• esa-sceva/satcom-chat-8b – Fine-tuned LLaMA 3.1 8B model for SatCom and scientific Q&A
• esa-sceva/satcom-chat-70b – Large-scale domain-specialised model (LLama 3.3-70B-Instruct) for high-level reasoning and expert assistance

Datasets

• esa-sceva/satcom-qa – Expert-authored QA dataset covering SatCom operations and systems
• esa-sceva/satcom-mcqa – Benchmark dataset for multiple-choice evaluation
• esa-sceva/satcom-synth-qa – Synthetic QA data for scaling domain coverage
• esa-sceva/satcom-synth-qa-cot – Chain-of-thought dataset for structured reasoning

🧩 Technical Highlights

• Based on Llama 3.1 and LLama-3.3 architectures (8B and 70B)
• Trained using Lit-GPT (Lightning AI) and EuroHPC infrastructure
• Incorporates Retrieval-Augmented Generation (RAG) for document-grounded factuality
• Includes a growing evaluation suite for link budget, RF design, and SatCom workflow analysis

🔓 Open Science Commitment

ESA-SCEVA follows the open science and open-source principles defined by the ESA ARTES programme.
All released assets — including models, datasets, and benchmarks — are:

• Freely accessible under permissive licenses (Apache 2.0 or equivalent)
• Fully documented for reproducibility
• Developed in collaboration with European research and industrial partners

🤝 Consortium

The SCEVA project is carried out by:

• Pi School – Lead partner and AI development
• RINA Consulting – Systems engineering and SatCom expertise
• European Space Agency (ESA) – Sponsor and technical authority under ARTES Connectivity & Secure Communications

ESA-SCEVA: advancing open, reliable, and domain-specialised AI for Europe’s space communications sector.