From Prompt to Public Service: Putting Advanced AI Use Cases into Production for Government Teams

This was the final session in CoPPER’s inaugural AI workshop series — four Thursday sessions in July 2025 where state employees demonstrated how they’re actually using AI in their work. The other sessions covered AI policy overview (NJ Office of Innovation), Power Automate integration (NJ DCF), and qualitative data analysis. I closed it out with the production-focused session.

What the workshop covered

The framing was simple: getting AI out of the demo stage and into real systems. A lot of state teams have played with ChatGPT or run a few prompts, but going from “this is cool” to something your agency actually depends on is a different problem. That’s what I tried to address.

I demonstrated three areas where I’ve done this work directly.

Vibe-coding websites. Using generative AI to draft a working landing page in minutes — not as a parlor trick, but as something program teams can use when they need a public information site quickly. I showed how to maintain state branding and ADA compliance in the output, and where you still need a human checking the work before anything goes live.

Regulation-aware chatbots. Staff spend a lot of time answering the same policy questions. The solution I showed uses retrieval-augmented generation (RAG) to query an agency’s own policy documents and regulations, returning answers with citations. The citation piece is non-negotiable for government — you need to know where an answer came from, and staff need to be able to verify it.

Guardrails and infrastructure. This is the part people skip when they’re excited about a demo: credential management, Azure OpenAI versus AWS Bedrock, data residency, audit logging, what happens when a security team reviews your new tool. I covered the practical mechanics of keeping experiments inside compliance boundaries.

Series context

The full series ran Thursdays in July 2025, each session 1.5 hours (1 hour of demo and discussion, 30 minutes of Q&A). The format worked well — state employees asking each other questions about real problems they’re dealing with, not a vendor pitch.

The Q&A for this session ran long, which I took as a good sign. Most questions were about procurement constraints and how to get security teams on board early, which are the right questions to be asking.