Two ITS developers built an AI agent that answers IT service ticket questions in the university’s own tone.

Tyson Mills, an enterprise software developer II, and Quan Chung, an enterprise full stack developer, were recently selected to participate in the AWS and MS AI Innovation Hub Summer Sandbox Slam for representatives from Mississippi state agencies. The Summer Slam is a learn-by-building experience where participants work in agency teams to design and build AI proof-of-concept prototypes using Amazon Quick Suite inside the AI Innovation Hub's AWS sandbox environment. Every Wednesday in July, they learned to use Amazon Q, which Tyson describes as AWS's counterpart to Microsoft Copilot.

Just as Copilot works with a client’s files and messages in the Microsoft environment, Amazon Q operates securely within the AWS ecosystem.

After an introduction to Q, each of the university teams began building a proof of concept for a solution to an existing challenge at their institution. As Tyson and Quan considered ideas that could be helpful at MSU, they thought about an agent in Copilot that Tyson had already built. The tool serves as a response playbook for ITS service tickets and provides guidance when staff encounter issues they do not handle frequently. The agent can search all ITS tickets by keywords, initiator, responsible party, dates, etc. and can also forecast trends, give summaries, and identify gaps in service. They began building a similar agent in AWS Quick, with the goal of learning transferable skills that would strengthen their Copilot tool.

The bootcamp required teams to use nonproprietary data that would not be linked to any university. To meet the bootcamp requirement, Tyson used AWS tools to generate a fictional institution called North Ridge University, complete with thousands of background files, invoices, purchase orders, student records, class schedules, and ITS help desk tickets. That material became the agent’s knowledge source. Besides answering questions, the agent the team built at the bootcamp was able to generate graphs, trends, and ticket summaries, and that gave Tyson and Quan ideas for what their own tool could do.

Tyson returned to MSU motivated to improve the agent he had created in Microsoft Copilot. One improvement involved incorporating the MSU tone agent that Quan had created, which transforms documents into a style and tone aligned with OPA guidelines. Quan and Tyson decided to have the two agents link to each other so that when Tyson’s agent responded to a question about an ITS procedure, it would respond in MSU tone, voice, and style.

Tyson found that having the ticket agent use the tone agent for a response slowed down the process. He learned that using workflows is more efficient than using subagents. “It’s better to have more workflows and fewer agents,” he explains. “I want to make the agents smart and have the workflows do the one specific task they are supposed to do. For example, one workflow will get the specific ticket details by ID, and another workflow will return all tickets in the last month. The agent takes the data provided by the workflows and synthesizes the answer. "We want the agents to be smart and the workflows to remain focused on a single task," Tyson explains. They simply need to return knowledge the agent can use.”

The team took the knowledge base and history of the MSU tone agent Quan had created and updated the instructions of the ticket response agent to say, “Whenever responding, use the MSU tone and playbook.” On the last day of the bootcamp, Tyson presented the MSU team’s project to the class, and it was among five projects the class chose to present to the Jackson Metro AWS Meetup, a broader audience in the Jackson metro area.

The team is beginning to consider other applications for the project. They would eventually like to make the agent available across campus so that troubleshooting suggestions for frequently reported IT problems are accessible to people when they need them, saving time for both users and ITS staff. The experience reinforced how rapidly AI tools are evolving and how skills learned on one platform can often be adapted to another.