Cheadle Center for Biodiversity and Ecological Restoration

Although taxonomists have access to a massive amount of intormation, accessing this data 1s a major hassle lor researchers and can be a barrier to entry for aspiring taxonomists. We created a chatbot for bee research that stores some of this information (provided by our advisors) and uses it to respond to user queries. Our approach was to implement a multimodal Retrieval Augmented Generation (RAG) pipeline, which matches user queries with relevant context in our bee data, returning answers based on related text chunks, along with mages. This method yielded great results, with our advisors evaluating the chatbot highly compared to other prominent Large Language Models (LLMs). We are hosting the model as a website locally, and plan publish it in coordination with our advisors after more rigorous evaluation of its performance. This poster was created as a product of the 24-25 UCSB PSTAT capstone (PSTAT 197).