Technical Consultants charging by the hour like lawyers are almost non-existent in the market - Mizunara Inc. Case Study
Mizunara Inc. successfully implemented a disaster prevention document platform for a major municipality by leveraging Leach's Generative AI consulting. They optimized vector search design for a 60-million-yen project, achieving stable operation with zero maintenance overhead.
📋 Article Processing Timeline
- 📰 Published: April 28, 2026 at 18:50
- 🔍 Collected: April 28, 2026 at 10:01
- 🤖 AI Analyzed: April 28, 2026 at 10:04 (3 min after Collected)
A disaster prevention document platform for one of Japan's largest local governments — how to achieve 'similarity search' for thousands of pages in a project worth approximately 60 million yen under a 3-year contract. It's not worth hiring a full-time architect, but the company couldn't decide on its own. Mizunara Inc. utilized Leach's Generative AI Consultant to fill this gap. We spoke with CTO Masayuki Oda and CEO Yuma Kinoshita.
'There are almost no people in the market who can act as technical consultants and charge by the hour like lawyers,' says Masayuki Oda, CTO of Mizunara Inc.
*The disaster prevention initiatives introduced in this article include projects conducted under Mizunara's predecessor system. The mentioned figure of approximately 60 million yen indicates the scale of the entire initiative (including infrastructure) and does not refer to Leach's consulting fees or maintenance contract amounts. For privacy reasons, client names and some specific information are withheld.
Support Summary:
- Approx. 60 million yen: The scale of the disaster prevention project for one of Japan's largest local governments (3-year contract, including infrastructure).
- 7 candidates to 1 plan: Comparison of MySQL full-text search, Aurora PostgreSQL + pgVector, etc., based on cost, operational load, and requirement suitability.
- Near-zero maintenance: Similarity search has been used daily in-house for about a year since launch, with operations remaining stable.
- Completed with 1 PDF: The design was finalized with time remaining on the contract, to the point where they felt 'bad about asking for more.'
Corporate Profile:
- Company Name: Mizunara Inc.
- Characteristics: A university-launched venture recognized by the Advanced Institute of Industrial Technology (AIIT).
- Main Initiatives: 'Wordless' for large-scale document management, 'OffiStill' office tool for the AI era.
- Structure: 7 members (2 full-time).
- Theme: Design consulting for similarity search in a disaster prevention document platform worth approx. 60 million yen.
- Web: https://www.mizunara.io/
1. What is Mizunara? — Building the foundation for handling large-scale documents.
Mizunara focuses on how to manage, edit, and publish large-scale documents. They are researching and developing 'Wordless' and 'OffiStill'. Real business documents aren't just Markdown; they need mechanisms to handle table data, diagrams, and citations at a 'business-ready granularity'. This project was an extension of that.
2. Challenges before introduction — Similarity search is needed, but technology cannot be decided due to specification and budget constraints.
The project was a disaster prevention document platform for a major municipality. It involved storing information texturized via OCR from massive PDFs/images and making it searchable on the web. The biggest issue was the inability to decide which technology to use for 'finding similar descriptions'. They were at a loss between existing MySQL search limits and unseen solutions. Leach suggested vector search, a path Mizunara couldn't have reached alone.
3. The '100% Achievement' constraint unique to municipal projects.
In municipal projects, following the specifications is absolute to avoid losing subsidies or disrupting the Gantt chart. They needed technology that 'fits the constraints' rather than just being 'good technology'.
'There are almost no people in the market who can act as technical consultants and charge by the hour like lawyers,' says Masayuki Oda, CTO of Mizunara Inc.
*The disaster prevention initiatives introduced in this article include projects conducted under Mizunara's predecessor system. The mentioned figure of approximately 60 million yen indicates the scale of the entire initiative (including infrastructure) and does not refer to Leach's consulting fees or maintenance contract amounts. For privacy reasons, client names and some specific information are withheld.
Support Summary:
- Approx. 60 million yen: The scale of the disaster prevention project for one of Japan's largest local governments (3-year contract, including infrastructure).
- 7 candidates to 1 plan: Comparison of MySQL full-text search, Aurora PostgreSQL + pgVector, etc., based on cost, operational load, and requirement suitability.
- Near-zero maintenance: Similarity search has been used daily in-house for about a year since launch, with operations remaining stable.
- Completed with 1 PDF: The design was finalized with time remaining on the contract, to the point where they felt 'bad about asking for more.'
Corporate Profile:
- Company Name: Mizunara Inc.
- Characteristics: A university-launched venture recognized by the Advanced Institute of Industrial Technology (AIIT).
- Main Initiatives: 'Wordless' for large-scale document management, 'OffiStill' office tool for the AI era.
- Structure: 7 members (2 full-time).
- Theme: Design consulting for similarity search in a disaster prevention document platform worth approx. 60 million yen.
- Web: https://www.mizunara.io/
1. What is Mizunara? — Building the foundation for handling large-scale documents.
Mizunara focuses on how to manage, edit, and publish large-scale documents. They are researching and developing 'Wordless' and 'OffiStill'. Real business documents aren't just Markdown; they need mechanisms to handle table data, diagrams, and citations at a 'business-ready granularity'. This project was an extension of that.
2. Challenges before introduction — Similarity search is needed, but technology cannot be decided due to specification and budget constraints.
The project was a disaster prevention document platform for a major municipality. It involved storing information texturized via OCR from massive PDFs/images and making it searchable on the web. The biggest issue was the inability to decide which technology to use for 'finding similar descriptions'. They were at a loss between existing MySQL search limits and unseen solutions. Leach suggested vector search, a path Mizunara couldn't have reached alone.
3. The '100% Achievement' constraint unique to municipal projects.
In municipal projects, following the specifications is absolute to avoid losing subsidies or disrupting the Gantt chart. They needed technology that 'fits the constraints' rather than just being 'good technology'.