Thanks to Melanie McDonough for continuing to tangibly show how AI can meaningfully impact the work of local governments!
Chief Innovation and AI Officer @ City of Lebanon, New Hampshire | Driving Innovation and AI Integration
Finally had some time to work on my Zoner agent (https://lnkd.in/ejiFFWyK). For those who have been following along, I used OpenAI's agent builder to begin working with messy data from the outside in. When it comes to innovation, my mantra has always been, "if you build it, they will come". Trying to get people to change the way they behave when they see absolutely no measurable value for themselves extends the time, energy, and sanity needed to arrive at the desired destination. Rather than try to convince overworked staff that they need to change their delivery of information (aka. generate structured data), I'm using AI to build a prototype for the future that will benefit them from day one. The idea is that when they see it, they will want it and will work collaboratively to build it better. It's worked for me in the past so much so, that no one really remembers how we go to certain versions of modernization that we are at today. In other words: show, not tell. My Zoner v1 agent was a good start and showed how to take our 200+ page Zoning Ordinance and deliver it as a focused chatbot that could answer questions related to our zoning ordinance. I learned a lot. This is not a skill I possessed out of the box. OpenAIs agent builder made it easy. However, every person who tested it for me asked the same thing. The chatbot couldn't answer based on address. So, I went back to work learning how to build an MCP that could talk to a GIS endpoint to look up zoning district by street address. That took me a little longer but today I pushed version two of my Zoner agent that is able to talk to my ZonerMCP and return answers specific to addresses in our city. There are a few things I need to smooth out but I finally have it working and learned a lot in the process. I also now have an MCP that I can use for other agents unrelated to this one. But more importantly, I have a basic prototype I can use to begin collaborating with our zoning administrators to deliver a tool at very low cost that actually helps solve real world problems. The agent takes in user feedback, runs it through a guardrail node to check for malicious use, determines if it is a zoning question, if so, determines if there is enough information to begin a search. If not, asks for zoning district or street address. If zoning district is provided, it heads off to perform the search. If street address is provided, it normalizes the address then uses the mcp service to look up the zoning district and runs off to perform the search. There are a few more things going on but that's the basic gist. I need to work out a few known bugs and update the UI and add some wizardry widgetzry but it's working for prototype purposes. And since it's built in agent builder, the possibilities are somewhat limitless. The main limits will be on data cleanliness which now, there is tangible evidence for making the effort. #AIforSocialGood #CivicTech #OpenAI #AgentBuilder