Fujitsu introduces generative AI service for source code analysis and design document generation

Fujitsu has announced the launch of Fujitsu Application Transform powered by Fujitsu Kozuchi, a new generative AI service designed to analyze source code and automatically generate associated design documents. According to the company, the service can reduce the time required for design document creation by up to 97%, thereby accelerating the modernization of legacy systems.

The service utilizes Fujitsu's proprietary technology combined with generative AI to analyze source code, including COBOL, from existing legacy systems. The aim is to produce easily understandable design documents that facilitate system comprehension and modernization efforts. Fujitsu claims that its approach, which incorporates Retrieval-Augmented Generation (RAG) linked to a knowledge graph, minimizes inaccuracies and omissions that can occur with general AI models, leading to a 95% improvement in completeness.

This new offering addresses the significant challenge organizations face when modernizing their IT infrastructure, where understanding older systems and their documentation is often a time-consuming process requiring specialized expertise. Fujitsu suggests that the service lowers the barrier to entry, enabling more efficient documentation creation even for personnel without deep knowledge of specific programming languages. The readability of the generated documents is reported to have improved by 60% compared to conventional methods.

Fujitsu will begin offering the service as Software as a Service (SaaS) in Japan starting March 30, 2026. The company plans to expand the service's capabilities in the future to include support for rebuilding and automatically rewriting source code for future use. This initiative is intended to assist businesses in creating clear specifications of their current systems, thereby aiding in the development and execution of modernization and migration strategies.