Fujitsu launches a generative AI application modernization service that analyzes legacy systems and automatically generates source code documentation.

Fujitsu announced the launch of its generative AI platform Fujitsu Kozuchi (Fujitsu Kozuchi), Fujitsu Application Transformation (Fujitsu Application Transformation Service). The service uses generative AI to automatically analyze the source code of legacy systems such as COBOL and produce design documents, greatly simplifying the extremely time-consuming manual analysis process of the past, helping companies improve the efficiency of modernization and transformation. Effective immediately, the service will officially be offered to enterprises worldwide in a SaaS model.

Fujitsu’s generative AI platform Fujitsu Kozuchi is focused on reducing development time

Fujitsu’s new service is mainly aimed at legacy systems (Legacy System) that are difficult to maintain within enterprises. It uses a generative AI platform called Fujitsu Kozuchi (Fujitsu Kozuchi) to automate the code parsing and document writing processes that previously had to be done manually by engineers. According to data provided by the company, the service can reduce the time required to generate design documents by about 97%. Compared with handling purely manually in the past or using general-purpose generative AI, this technology addresses the problem that legacy programming languages (such as COBOL) have high barriers to understanding and are time-consuming by means of a standardized analysis process. It enables enterprises to grasp the existing systems’ architecture and characteristics more quickly, laying the groundwork for subsequent system migration or modernization and transformation.

How does it solve the problem of AI hallucinations?

To ensure that the produced technical documents are accurate, Fujitsu combines knowledge graph (Knowledge Graph) and retrieval-augmented generation (Retrieval-Augmented Generation, RAG) technologies. This RAG management mechanism, optimized for software engineering, can precisely extract related information from a large amount of source code, avoiding the common hallucination risk of generative AI. Technical data shows that this approach can increase the completeness of information to 95% and prevent information omissions. By building an asset knowledge graph and collaborating with large language models (LLM), the readability of the resulting design documents is improved by 60% compared with traditional methods, effectively solving the code-understanding challenges under complex logic.

What specific features does software analysis and visualization have?

The service is divided into two core components: first is “software analysis and visualization.” This function can automatically draw software maps, presenting the functional structure of applications in an intuitive way. It classifies all programming languages, identifies redundant or similar source code, and thereby evaluates the difficulty level of migrating the system to an open environment. Second is “reverse engineering of software design documents.” This part focuses on extracting logic from existing assets and generating easy-to-read documents. With this integrated approach, enterprises can not only visualize the structure of their black-box applications, but also obtain high-quality technical documentation, which has significant practical value for industries such as finance and retail that lack original design data or face gaps in development personnel.

Fujitsu plans to supply this technology in Japan starting March 30, 2026 via software-as-a-service (SaaS), and expects to gradually expand functionality starting in fiscal year 2026. Future development directions include refactoring existing source code (Refactoring), automatically rewriting code, and ongoing operational and maintenance support. In addition, the R&D team also plans to introduce interactive features so developers can communicate with AI and verify application specifications or assess the impact of code changes.

This article Fujitsu launches generative AI application transformation service, automatically generating source code documents for legacy systems first appeared on Lianxun ABMedia.