Creating a CREWES AI chat assistant using gpt-4o-mini and retrieval-augmented generation

Marcelo Guarido, Kristopher A. Innanen

Large language models, such as the GPT or Llama models, are popular tools that can be used for many applications. We developed the CREWES AI Chat Assistant powered by gpt-4o-mini using retrieval-augmented generation. The chat assistant answers geophysical questions covered by the CREWES research scope. The app is fed with the CREWES research reports from 1989 to 2023 and was able to answer related questions with deep explanations, using equations and returning the source file as a reference. It also has multi-language capabilities, answering questions in all the languages the model supports. Finally, any corporation can modify the algorithm to work with its internal files and create its own AI Chat Assistant.