9. Conclusion
Jupyter Notebooks are widely in use by the scientific community to run experiments and analysis. Though their user-friendly interface is intended to make coding more accessible for researchers in fields outside computation, using application-specific APIs can still be a challenge for many. In this paper, we have presented an overview of a RAG-based AI tool that we built to help users of the NSF FABRIC network testbed. Our results show RAG approaches, in which the user’s query is augmented by a vector database of relevant code examples published by the FABRIC development team, combined with pre-trained LLMs can produce usable or nearly usable code for users in most cases. As our tool is built with open-source libraries and uses relatively small LLMs that can be run locally, similar tools can be developed easily for other scientific software even by a small group.