Retrieval Augmented Generation (RAG) for Space Mission Design - A Space Mission Design Assistant

Abstract

​​Early-phase space mission design is characterised by vast design tradespaces and significant uncertainty, where effective retrieval of historical knowledge is crucial yet often inefficient, even with the adoption of Model-Based Systems Engineering (MBSE). This paper presents a novel Retrieval-Augmented Generation (RAG) framework to streamline this process by leveraging a Large Language Model (LLM) as an intelligent design assistant. The system integrates a curated knowledge base derived from the European Space Agency (ESA)’s eoPortal archive with a LlamaIndex-based RAG pipeline. The methodology involved developing a high-throughput scraper (achieving a >4×speedup and 99.9% data acquisition success), robust data preprocessing (including Markdown-based structured data conversion), and a comprehensive evaluation via a parameter sweep against a synthetically generated benchmark. Results demonstrate a highly effective system with high retrieval accuracy (Mean Average Precision (MAP@k) ≈0.95) and strong generation quality (F1-score ≈0.75). The optimal configuration (top_k=5, similarity_threshold=0.5, temperature=0.1) provides faithful and relevant answers to complex technical queries within acceptable latency (median 2.0 seconds), verified through qualitative analysis of a Streamlit prototype. This framework significantly enhances data-driven decision-making in early-phase space mission analysis, laying a robust foundation for future, deeper integration with MBSE workflows.​