Improving the chatbot, Chatür, at the University of Arizona

April 12, 2024

Large Language Models (LLMs), a remarkable technological advancement, have garnered significant attention and interest among users in academia and beyond. The access to these LLM platforms is commonly facilitated through chatbot-like interfaces provided by various organizations, such as the widely known OpenAI's ChatGPT. These organizations often employ a freemium model, where the fundamental version (e.g., ChatGPT 3.5) is available free of charge, while advanced features and enhanced quality require a paid subscription.

In the realm of education, artificial intelligence chatbots like ChatGPT have become increasingly pervasive. Students leverage these chatbots for a variety of academic pursuits, such as completing coursework, grasping concepts, and conducting research. On the other hand, instructors utilize AI chatbots to craft instructional content and assist with grading tasks. Researchers have also embraced these systems as valuable tools for lead discovery and data synthesis.

Generative AI technology like LLMs presents a common challenge known as hallucinations. These systems sometimes produce false or fabricated information that can be difficult to identify, especially if one is not familiar with the topic. To reduce the frequency of hallucinations, a technique called Retrieval Augmented Generation (RAG) is often used. This method restricts the system's response to information provided to it, rather than allowing it to generate content from scratch. Realizing these challenges and demands, the Data Science Institute (DSI) and Institute for Computation and Data-Enabled Insight (ICDI) were tasked in January 2024 to explore how LLMs could be used to support the needs of five unique use cases proposed by UArizona educators and research projects. About 30 participants of the three-day codefest created a secure chatbot called Chatür that used open LLMs.

A subset of the team continued working on Chatür and released a stable version for testing to stakeholders on March 1, 2024. The feedback from stakeholders motivated DSI and ICDI to improve Chatür in accuracy and scalability. Their goal is to integrate Chatür with systems like the D2L learning system at UArizona, making it easily accessible to instructors and students.

By August 1, 2024, the team plans to deliver an enhanced version of Chatür (Version 2) that can generate precise responses to queries. This improved chatbot will be validated by four stakeholder groups (professors, students, researchers, and administrators) to ensure its effectiveness in supporting the UArizona learning system.

In September 2024, the team plans to expand the infrastructure to accommodate research projects, further enhancing Chatür's capabilities and contributions to the UArizona community. Email the Data Science Institute if you want more information about Chatür at

Tina L. Johnson