New Jersey releases upgraded AI assistant for state employees

Three technologists with the New Jersey Innovation Authority on Tuesday announced major upgrades to the state’s in-house generative AI assistant — a “2.0 version” of the tool.

The New Jersey Innovation Authority, which was previously known as the New Jersey Office of Innovation before former Gov. Phil Murphy codified the office into law with a new name in January, originally launched the state’s internal AI assistant in 2024. News of the updates — which include more transparent responses, conversation control and built-in trainings — was shared in a blog post on the state’s website, coauthored by Katherine Nammacher, product lead of the platform team; Oreofe Aderibigbe, product designer of the platform team; and A.J. Hyppolite, a software engineer.

The three shared that when the tool launched two years ago, they worried about its reception and the lack of precedent of a proprietary generative AI assistant for state government.

“300,000 sessions and 1 million prompts later, we now know that the demand was and is real,” the post reads. “Every week, thousands of professionals in our 70,000-member state workforce use the tool to draft and edit, summarize long documents, brainstorm new approaches, and more. In recent months, we’ve surveyed hundreds of them and learned more about the cutting-edge features they want and expect from the tool. We also did in-depth interviews to understand exactly how people were using AI, and what they needed moving forward.”

The responses led the trio to “rebuild the tool from the ground up,” creating second version, with new features. Those changes included more transparency, allowing users to see the tool’s reasoning through a “thoughts” section, a feature found on commercial chatbots like ChatGPT.

“That’s important because our government colleagues often need to know how the tool arrived at its output, helping them to catch errors, verify logic, identify possible bias, and make informed judgments about whether and how to use the tool’s response in any given situation,” their post read. “In other words, we’ve made it easier to check work and spot any hallucinations — confident-sounding replies from the tool that might actually not be correct.”

Users will also be allowed more control over input — new features include the ability to speak a prompt instead of typing it, to edit a prompt instead of starting over, retry a question when they’re not satisfied with the first response and to view conversation history.

From directly within the tool’s interface, state employees can access guides, such as a guide on how to draft effective prompts or the state’s acceptable use policies. The developers said the tool, like the previous version, is still hosted on the state’s network, ensuring state data doesn’t leave the environment, keeping safety and security a top priority.

One of “the biggest long-term values” of the upgrades, they said, is that it’s faster to add new tools — making it easier for the chatbot to be reflective of the needs of New Jersey’s government and its employees.

“Notably, this is possible because we used an open-source software [large language model] interface that allows us to quickly incorporate new features while still maintaining control of those features, capabilities, and implementation,” the post read. “Overall, we envision this tool to remain the safe, go-to tool for State employees, helping them learn and practice responsible AI use while getting things done better and faster for New Jerseyans.”