NASA NTRSDate not supplied
NASA has a vast archive of engineering guidelines, standards, and best practices collected over decades. This encompasses a breadth of topics from rocketry and engineering standards to risk management and space-related health issues. This wealth of information, while invaluable to NASA engineers, staff, and the public, is too extensive for any individual to fully comprehend. To address this challenge, we have developed Atlas, a tool within NASA's Mission Cloud Platform that enables users to query these diverse sources effectively. Atlas allows users to ask natural language questions and receive answers grounded in factual information from source documents. The tool provides responses with direct quotations and links to original documents, ensuring transparency and accuracy. It can address a wide range of queries, from specific technical details like safe distances for rocket launches from lightning to broader topics such as crew health requirements for long-duration space missions, corrosion protection in low Earth orbit, and NASA's agreements with various entities. In developing Atlas, we encountered and overcame several technical challenges. Large Language Models often struggle with consistently providing accurate information, especially for highly specialized topics. We implemented strategies to prevent hallucinations and ensure the reliability of responses, even for complex questions on topics ranging from NASA Mission Classes to intricate rocket science concepts. Additionally, we addressed the challenges of delivering quick responses while maintaining cost-effectiveness. Our presentation will detail the innovative approaches we employed to optimize performance and efficiency, making Atlas a powerful and practical tool for accessing NASA's extensive knowledge base.