Legacy systems don’t fail because they’re old. They fail because nobody understands them anymore. This is a problem that quietly costs enterprises billions every year.
At the Tech: Europe London AI Hackathon, my partner Maxime and I built Legacy Atlas, a tool that acts as an archaeologist for legacy codebases. You give it a codebase, and it reconstructs the full picture: the business logic, execution flows, integrations, risk scores, and safe migration advice. An agentic layer lets developers ask questions like “what breaks if I change this?”
The key architectural decision was to run deterministic analysis first (AST parsing, call graphs, entity recognition) and layer AI interpretation on top, powered by CodeWords automations, Dust agents, and the Gemini API.
We reached the finals of the Conduct AI track, competing against 15 teams and over 100 participants. Built the whole thing in a single day with Dust, Google DeepMind, and Agemo as technology partners.
Demo