AI Bright Spot: Lowering costs and modeling accountability with AI in Vermont

The Partnership for Public Service AI Center for Government® is publishing a series of blogs to celebrate how artificial intelligence and intelligent automation are being used by the government to serve the public.

We recently spoke with Rose Vath, AI & business intelligence and performance manager at the Vermont Agency of Transportation. Vath, an alumna of the Partnership’s AI Government Leadership Program, shared insights on the state government’s internal ChatVT AI assistant and her agency’s work on an AI-powered knowledge repository.

A paced approach

From her position in the Vermont Agency of Transportation, or AOT, Rose Vath credits the state government’s centralized IT service as the basis for the state’s early leadership on AI strategy. “We were the first state to have a chief AI officer in state government, and he sits in our Agency of Digital Service,” she explained.

As manager for the AOT’s business intelligence unit, Vath’s focus is determining what opportunities exist for her agency to adopt AI tools and the best tools to service those opportunities. She said her team has only recently started to lean into tools offered through Microsoft Enterprise, like AI Hub available through Power Automate and Copilot Studio.

Vath cites a clear reason for this paced approach: a desire to proactively reduce the risks that could come with broad-scale AI adoption. “Injecting one use of AI into a deterministic workflow is far less risky than transitioning a workflow to an agentic AI solution, where we can’t control the inputs or outputs,” she said.

Lowering costs with ChatVT

Alongside external AI tools, Vermont’s government is examining opportunities to develop in-house AI resources. One of the most prominent use cases is ChatVT, an internal AI assistant for Vermont’s state employees.

As Vath explains, the inspiration for ChatVT originated from the flexibility afforded to Vermont’s employees to make use of external AI applications: “We’re a unique state in that our policy allows us to use publicly available AI tools, provided we’re working with public data that can be used appropriately with those tools. So, our state employees can use things like ChatGPT, Claude and Gemini.”

At the same time, using these external AI tools posed significant potential cost for the state government, particularly for the tokens associated with queries. “ChatVT is our safety hatch if these [tools] start to become financially burdensome,” Vath said. “With ChatVT, we have a tool that we had the opportunity to design with the express goal of cost minimization.”

Across the state government, ChatVT now has about 500 weekly users, Vath said. While ChatVT is still an internal state government service, Vath explained how a public deployment of ChatVT could pass savings currently seen by the government to taxpayers. With a model like ChatVT’s “people [won’t] have to pay license fees for premium services, and we think that would be especially helpful for small businesses,” she said.

Storing specialized knowledge

While ChatVT represents a general use of AI for individuals, Vath’s team at the Agency of Transportation has also investigated more specialized uses of the technology to support enterprise applications.

One of the specialized use cases that Vath’s team is developing is an AI-powered knowledge repository specific to the Agency of Transportation. This repository would store and support queries for about 75 different procedures, processes, manuals and handbooks that are currently maintained by a variety of teams within Vath’s agency.

“It’s a perfect use case for AI’s natural language processing,” Vath said. “A new project manager could come in and query this tool to understand more of our processes instead of navigating a fragmented information environment.”

Vath has refined several aspects of this planned use case during her time in the AI Government Leadership Program. “One of the things that came out of the program was how to begin to think about the evaluation,” Vath said. “It’s not like traditional software where you write a script and click through buttons.”

Based on what she learned in the program, Vath has now planned a robust evaluation process for the knowledge repository. This process will include developing a library of test prompts followed by test users grading the completeness and accuracy of the AI’s responses to these prompts.

Training and accountability

Beyond evaluations, Vath also sees training and support for state employees as critical to the success of the knowledge repository and other AI use cases. Vath leads a monthly AI training workshop for the Agency of Transportation called On Ramp for AI at AOT. In this workshop, Vath covers topics ranging from “What is AI?” to scenario-based walkthroughs of services like ChatVT and other approved AI applications.

In leading these training sessions, Vath sees herself as providing an invaluable source of support to her team: “In Vermont, one of our core tenets is accountability. I see myself as accountable for how AI is implemented and used at my agency. I want our staff to know that if something goes wrong, they will have support. That’s why the workshop is important.”

---

Continuing the conversation

The AI Center for Government champions AI innovators across all levels of government. If your agency is taking steps to lead AI well, we would love to hear from you. Join us as we highlight real-world AI use cases and convene public sector leaders from across the country to share tools and insights to lead confidently in the age of AI.

We’re here to help!

• Sign up for our newsletter.
• Follow us on LinkedIn.
• Get in touch! Email us at [email protected].