From Spaghetti to Structure: Decoding Complex Codebases with Graphs and GenAI
summary
Claim: lack of documentation blocks modernization for 96% of customers and 41% report productivity loss. Built Decipher, a pipeline that replaces regex with AST extraction, maps methods/classes and their interactions into a Neo4j graph, runs Cypher to find entry points and call-stack-based processes, enriches nodes with LLM summaries, auto-generates per-node docs, sequence diagrams and domain clusters. Initial naive LLM summaries fail on large repos but the graph + GenAI approach scales and produces usable design docs.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 2:10Legacy code challenges
- 6:30Context window limitations
- 9:42Graph database breakthrough
- 13:14Process and domain extraction
- 15:44Domain-level documentation
- 17:32Future of decipher tool
- 19:58Summary and thanks
related talks
08mHarness Graph Power With Agentic AI: An Insights Discovery Use Case
40mAI’s Big Leap
31mBeyond Vectors: Evolving GenAI through Transformative Tools and Methods
01h 00mGraph Mastery to AI
36mGraphRAG Explained 🔥 Building Smarter LLMs
01h 17mGraph Database and Retrieval Augmented Generation
22mBuilding Cypher Queries With Open-Source LLMs
27mGraph-Powered Storyworlds: Using Neo4j To Keep 1M+ Word LitRPG Epics Coherent w/ AI