GenAI for Legacy Systems Modernization

While most people actively write about using GenAI tools to generate new code, there is a new Thoughtworks publication that focuses on the opposite — using AI to understand and refactor legacy systems.

What makes legacy systems modernization expensive?
- Lack of design and implementation details knowledge
- Lack of actual documentation
- Lack of automated tests
- Absence of human experts
- Difficulty to measure the impact of the change

To address these challenges Thoughtworks team developed a tool called CodeConcise. But the authors highlighted that you don't need exactly this tool, the approach and ideas can be used as a reference to implement your own solution.

Key concepts:
✏️ Treat code as data
✏️ Build Abstract Syntax Trees (ASTs) to identify entities and relationships in the code
✏️ Store these ASTs in graph database (neo4j)
✏️ Use a comprehension pipeline that traverses the graph using multiple algorithms, such as Depth-first Search with backtracking in post-order traversal, to enrich the graph with LLM-generated explanations at various depths (e.g. methods, classes, packages)
✏️ Integrate the enriched graph with a frontend application that implements Retrieval-Augmented Generation (RAG) approach
✏️ The RAG retrieval component pulls nodes relevant to the user’s prompt, while the LLM further traverses the graph to gather more information from their neighboring nodes to provide the LLM-generated explanations at various levels of abstraction
✏️ The same enrichment pipeline can be used to generate documentation for the existing system

For now the tool was tested with several clients to generate explanations for low-level legacy code. The next goal is to improve the model to provide answers at the higher level of abstraction, keeping in mind that it might not be directly possible by examining the code alone.

The work looks promising and could significantly reduce the time and cost of modernizing old systems (especially written on exotic languages like COBOL). It simplifies reverse-engineering and helps generate knowledge about the current system. The authors also promised to share results on improving the current model and provide more real life examples for the tool usage.

#news #engineering #ai