Standardizing Fragmented Public Data

Polimill identified that the primary barrier to AI adoption in the Japanese public sector was data fragmentation. Municipalities operated with disparate document formats and scattered historical records, making it impossible for AI to provide consistent policy insights. To solve this, Polimill created a high-precision search foundation by collecting and standardizing assembly minutes across the country, adding metadata to make administrative information searchable and usable as a unified knowledge base. This infrastructure now supports 1,050 municipalities and 550,000 public employees.

Accelerating Development and Adoption

By integrating OpenAI's Codex into their engineering workflow, Polimill increased development speed by 3-5x. The team utilized AI for the entire lifecycle, from requirements definition and code consistency checks to implementation and testing. This allowed engineers to shift their focus from manual coding to reviewing AI-generated plans and making high-level architectural decisions. Furthermore, the choice to build on GPT models lowered the barrier to entry for non-technical public employees, as the widespread familiarity with ChatGPT made the new platform feel accessible and reliable for daily administrative tasks.

Capturing Tacit Knowledge and Future Scaling

Beyond mere efficiency, Polimill is using AI to bridge the experience gap between veteran and junior officials. While AI-drafted policy proposals from junior staff performed well, they lacked the "tacit knowledge"—the nuanced, unwritten practical judgments—found in proposals from veteran officials. Polimill is now recording how these experts instruct and refine AI outputs to turn their implicit judgment into reusable organizational knowledge. Looking ahead, the company plans to launch "Qommons ONE" in late 2026, a platform featuring a "super agent" that can orchestrate multiple specialized AI systems and third-party apps to execute complex administrative workflows from start to finish.