Japan’s public sector has a data fragmentation problem that no amount of raw model capability can paper over. Polimill, a civic-tech company that started with a citizen engagement platform called Surfvote, realized that before you can meaningfully reflect citizen voices in policy, the machinery of local government itself needs to stop drowning in paperwork. So they built QommonsAI, launched in October 2024, and it’s already in use by roughly 1,050 municipalities and about 550,000 public employees across the country.
The platform covers assembly response drafting, public services, social welfare, and legal search. The real engineering challenge wasn’t the AI — it was the data. Each municipality in Japan has its own workflows, document formats, and scattered historical records. Polimill’s team collected and standardized assembly minutes from across the country, used AI to add metadata, and built a search foundation that works across municipalities and time periods. That’s the unglamorous work that makes generative AI actually useful in government, and it’s the piece most enterprise AI deployments skip.
Polimill runs on OpenAI’s GPT models. Masahiro Wakabayashi, the company’s CAIO, says the decision came down to broad capability and familiarity. Public employees may not know model names or technical jargon, but they know ChatGPT. That recognition lowers the adoption barrier when a tool gets rolled out to thousands of staff who didn’t ask for it. The platform also includes audit controls — administrators can review feature usage history and restrict which models are available, which matters when you’re dealing with public records.
The development story is just as interesting. Polimill adopted Codex across its entire workflow, and development speed jumped to 3-5x previous levels. Engineers now focus on reviewing AI-generated plans while the implementation work happens autonomously. Wakabayashi frames the validation cycle — prototype, show to municipal teams, gather feedback — as accelerating dramatically. OpenAI’s hands-on support, including guidance on where AI should handle steps versus where humans need to review, is something he considers as valuable as the API performance itself. The long-term vision is to make QommonsAI a common foundation that prevents service gaps between wealthy and struggling municipalities — a public OS for Japanese government, not just another productivity tool.
