An artificial intelligence powered zoning assistant may be added to Cleveland Township’s website pending a final board vote next month, possibly offering residents near instant answers to questions that usually require waiting for a callback.
“You are making it so there’s a 24-7 zoning agent,” Cleveland Township resident and software engineer Travis Kemp said. “You get an answer in three seconds versus waiting for two weeks.”
The tool, called ZoningOS, has been developed by local zoning administrator Scott Sheehan and Kemp under their company Blue Water Intelligence (BWI). The pair recently presented the idea to the Cleveland Township Board, offering to install the system on the township’s website at no cost as a pilot.
“If it’s something that can enhance our website and make it easier for users to navigate, then I’m all for it,” Cleveland township clerk Tracy Olsen said.
However, the rollout comes with some scrutiny, as Sheehan is both co-developer of the product and the township’s zoning administrator.
During a recent board discussion, supervisor Tim Stein raised the concern directly, saying, “I have a problem if we were going to be charged for it. I see that as a conflict of interest.”
Sheehan acknowledged the concern and said it factored into their decision to offer the tool to the township at no cost.
“To me, it just made sense to remove that question entirely,” he said. “I don’t want there to be any perception of a conflict of interest, so offering it to my own townships for free was the safest way to handle it.”
While Cleveland Township would not pay for the service, the ZoningOS contract outlines a subscription based pricing model for other municipalities, along with a separate onboarding fee of $250 and annual licensing terms, among other ranges of pricing, including how big the township is.
The idea grew out of a common challenge in local zoning offices: responding to frequent, often repetitive questions from residents, real estate agents, contractors, and developers, according to Sheehan.
That volume of questions can take a long time to answer, particularly in townships where officials often serve in part-time roles.
“What infrastructure exists to accommodate something being there to answer those questions in a timely fashion rather than waiting for me?” Sheehan asked.
The proposed small, oblong, teal colored widget is structured as a chat style feature embedded directly into the township’s website that’s for disability features.
Users can click the tool, type in a question about zoning or select from a pre-set list and receive an answer in seconds.
The system draws only from documents uploaded by the township, such as zoning ordinances, maps, etc. If it cannot find a clear answer, it directs users to contact the zoning administrator; the old fashioned way.
“We have it set to air on the side of caution…if it’s not clearly stated in their ordinance, then it will respond to contact the zoning administrator for clarification,” Sheehan said.
The proposed contract reinforces that it is intended for information use only and explicitly states that ZoningOS does not provide legal advice regarding zoning or land use decisions.
The system, which is an Enterprise API, that is using Anthropic and OpenAI models to help train is categorized as a dual-tiered large language model.
“We’re using a very advanced initial AI, a large language model (LLM). And then we have a second LLM from a totally different provider basically fact check the initial response. Only until both the first and second (models) agree do we even return an answer to the end user,” Kemp said.
Unlike broader AI tools that pull from the internet, ZoningOS operates within a closed set of information -- it can only access what documents and files the township uploads.
“The AI is looking at the ordinance and that’s it,” Sheehan said.
In addition to answering questions, the system tracks what users are asking in what Sheehan described as a “pulse of the township.”
That information is compiled into reports for township officials, identifying which questions are clearly addressed by existing ordinances and which may point to gaps or confusion.
Under the terms of the contract, that data can be retained for up to 730 days and may also be used in an aggregated, anonymized form for product development and benchmarking.
While BWI says it does not collect personal identifiers such as names or property ownership, the system still logs the content of resident inquiries which may include location details or development plans. Access to the data is limited to township administrators and designated personnel, according to the agreement, and all information is stored within U.S. based infrastructure.
Over time, the patterns could function as an informal feedback loop between residents and policymakers, though it is not yet clear to what extent the data will be used in decision making.
“What you cannot afford is to have the system give out bad zoning advice,” Kemp said. Sheehan emphasized the extent of testing conducted saying, “at this point, we can’t get it to break. We can’t get it to say the wrong thing.”
However, the contract also states that AI-generated responses may contain errors or omissions and should not be relied upon as legally authoritative determinations. Cleveland Township will be an early test of how it performs outside of controlled conditions.
More information on ZoningOS and Blue Water Intelligence can be found at zoningos. net.
