Author(s)
Makayla Kienlen, Bingyan Lu, and Nick Laneman
Abstract
Radio spectrum data from the Federal Communications Commission (FCC) is often complex, coming in many forms and contained in legacy systems, making it difficult to interpret and inaccessible to experts and non-experts alike. With the rapid advancement of Artificial Intelligence (AI) and Large Language Models (LLMs), these tools can be applied to spectrum data to make it more easily usable for researchers and the general public. In this project we created an AI-powered chatbot that can answer questions about structured, tabular spectrum data stored in SQL databases. The system uses Retrieval-Augmented Generation (RAG) to retrieve relevant data and provide it to the LLM as context, improving the accuracy of its responses. Additionally, we incorporate AI agents that perform dynamic, iterative processes to answer users’ questions, along with custom tools that help the LLM interact with tabular data. We also developed a comprehensive set of questions to evaluate the chatbot prototypes. The results show that using agentic RAG combined with domain-specific tools significantly improves the chatbot’s performance from 45% to 91% overall accuracy.