Streamlining NASA Research Retrieval with RAG-Inspired Semantic Search

The vast corpus of NASA-related research is dispersed across numerous isolated repositories, posing significant challenges for researchers in efficiently discovering relevant literature. This paper introduces a novel solution leveraging large language models and hybrid retrieval methodologies to streamline the search and retrieval process of NASA's extensive research publications. Our proposed system aggregates and scrapes abstracts, metadata, and other relevant details from disparate NASA-related publishing sources, ensuring structured and standardized indexing. As proof of concept for our proposed solution, we implement a combination of search techniques, including sparse retrieval (BM25), dense vector embeddings, and hybrid ranking via Reciprocal Rank Fusion (RRF). The system is designed to be modular, scalable, and computationally efficient, allowing for continuous enhancement and expansion. The resulting prototype is packaged as a Python-based Streamlit application to demonstrate its feasibility and accessibility. This tool aims to reduce barriers to literature discovery, foster multidisciplinary collaboration, and serve as a foundation for future improvements in AI-assisted research retrieval by enabling more precise and efficient access to NASA-related research.