GenAI-assisted reverse engineering for incremental Legacy modernisation
summary
Mainframe modernization uses incremental seams and GenAI to avoid big‑bang rewrites: a PoC called CodeConcise parses COBOL and Java with ASTs, builds a Neo4j knowledge graph, and combines RAG with LLMs to generate documentation and a chat interface. A client with ~15 million LOC sees reverse‑engineering drop from ~6 weeks per 10k LOC to ~2 weeks, and cloud forks (Kafka + KV store + replication) enable safe dual runs.
AI-written summary · about
chapters (14)
- 0:00Introduction to modernization
- 4:13Understanding mainframes
- 7:06Identifying mainframe seams
- 8:59Batch input seams
- 12:19Pipeline handoff seams
- 14:34Data characteristic seams
- 18:01Limitations of traditional seams
- 20:40GenAI for code discovery
- 23:29RAG for context awareness
- 25:46Knowledge graphs and LLMs
- 27:27ASTs and graph analysis
- 29:09CodeConcise solution model
- 33:46Results and future ideas
- 35:34Toolbox conclusion
related talks
16mAI-Powered Mainframe Modernization
41mIncremental Modernization through CodeConcise
43mAI accelerated legacy modernisation
36mRethinking legacy modernisation with automated reasoning and comprehension models
27mNavigating Application Modernisation - Leveraging Gen-AI
08mMigrating from COBOL with AI: A Moonshot Demo
28mApplication modernisation with generative AI
51mLegacy non-x86 modernization with