U-M Generative AI Use Cases

The Shared Services Center processes approximately 300,000 invoices annually, extracting key information from vendor-submitted vouchers. Previously, this process relied on a template-based system implemented in 2019, which was revolutionary at the time but had limitations, processing only about 30% of invoices. With the sunsetting of this older technology, a new AI-driven solution was needed to enhance efficiency, accuracy, and scalability in invoice processing.

New employees often have a wide range of questions during their onboarding process, from benefits and policies to campus resources and department-specific information. The challenge was providing timely, consistent, and easily accessible answers to these questions without overwhelming HR representatives and managers. U-M needed a solution that could centralize key onboarding information while ensuring new hires felt supported as they transitioned into their roles.

Using generally-available tools such as ChatGPT to write, debug, or modify Wolfram Language code can produce inferior results because these tools are not specifically trained to use Wolfram Language. Other Mathematica AI integrations similar to the one using U-M GPT Toolkit do not provide data and research privacy assurances, nor do they allow associated costs to be charged to a U-M Shortcode.

ACM (Association for Computing Machinery) students preparing for ICPC (International Collegiate Programming Contest) programming competitions and technical interviews often face challenges in finding relevant resources, receiving targeted feedback, and practicing in an interactive environment.

Websites need to meet accessibility standards and remove barriers for individuals using screen readers. Developers often spend significant time creating descriptive alt text for images, which is labor-intensive and prone to inconsistency.

Instead of searching through U-M’s standard practice guides (SPGs) on U-M’s HR website, ITS trained a Maizey to identify existing policies through plain text question prompting. The department’s goal was to simplify the process of searching for and accessing SPGs to save time and reduce the difficulty of finding relevant policies.

When it comes to transforming an engineering process from technical terms to logic, U-M GPT has become a helpful tool. U-M engineering and computer science students turned to the summarization ability of GenAI tools. This allows students to focus on the essence of design thinking as it relates to system building and problem-solving at a higher level.

Professor Anne Gere needed an efficient method to track down citations for her book, a task that can be time-consuming and tedious. Traditional methods of citation tracking were not sufficient to meet her needs for speed and accuracy.

The PMO needed to address several challenges, including the desire for a safer, more secure AI-driven solution for sensitive projects and the need for tools to aid in professional development. Team members, like Mallory Kwiatkowski, sought to enhance their skills and knowledge by leveraging outside resources and comparing various Generative AI (GenAI) tools, all while maintaining security and cost-effectiveness.

ITS Administration sought to efficiently rank and compare hundreds of resumes to streamline the hiring process and identify the most suitable candidates quickly.

Professor DeOrio sought to supplement existing course resources like office hours and discussion forums for his large EECS 485 Web Systems course, by providing quick feedback and assistance to students for one of their projects.

 Professor Jurgens wanted to change teaching practices to prepare students for the next generation in the era of AI. Specifically, Jurgens sought to help students understand AI not as a replacement for their work but as a tool for refinement, idea generation, critiques, feedback, and providing structure.