Rethinking legacy modernisation with automated reasoning and comprehension models
summary
Combines Gen AI with deterministic compiler techniques to ground reverse engineering: extract abstract syntax trees and control-flow graphs, run dominator and reducibility tests, and apply symbolic execution to produce incontrovertible facts. A Langraph agent builds hypothesis trees, invokes MCP-style analysis tools, and updates beliefs with Bayesian fusion (example posterior probability drops from 0.5 to 0.4), reducing LLM hallucination and giving traceable evidence back to code. Prototype includes a toy VM and working agent; approach scales via parallel deterministic analysis.
AI-written summary · about
chapters (11)
- 0:00Introduction to legacy code
- 1:18Reverse engineering artifacts
- 2:47Three pillars of modernization
- 5:09Structural analysis methods
- 8:54Graph theory for code analysis
- 14:23Parsing and robustness
- 17:35Symbolic execution
- 20:31Reasoning with Bayesian logic
- 23:27Validating AI hypotheses
- 28:16Context engineering and planning
- 32:32Conclusion: Deterministic foundation
related talks
46mDemystifying AI Agents
39mGenAI-assisted reverse engineering for incremental Legacy modernisation
24mAI Hates Legacy Code
50mFrom Prompt Hacking to Architectural Determinism: Engineering Reliable GenAI Systems
58mAdvanced Autonomous AI Agents: Memory, Collaboration and Evolution in Action
34mProgramming Languages for AI Programing Agents
01h 03mAgents Everywhere
35mRewilding Software Engineering - AI, Tools and Human Decisions