Download now

Executive Summary

This white paper shows how Google AI and CAST software mapping & intelligence combine to deliver deterministic, structure-aware understanding of large, custom applications – empowering LLM and making modernization decisions faster and safer.

Learn about:

  • How to scale LLM-powered understanding to million-line estates for reliable impact analysis.
  • An agentic approach that layers AI-generated documentation and vector search with CAST Imaging’s graph-based “digital twin” of the code to answer code-to-architecture questions with evidence.
  • Concrete gains for teams – from accurate dependency and data-flow mapping to tighter test scoping, risk detection, and defensible rehost/replatform/refactor choices.
Antoine LarmanjatGoogle Cloud
Antoine Larmanjat

Sr. Technical Director

Antoine Larmanjat is a Technical Director in Google Cloud’s Office of the CTO, leading EMEA engagements with a focus on AI for code and emerging tech. Previously Group CIO at Euler Hermes (Allianz Trade), he led a major IT transformation and drove an engineering culture built on microservices, APIs, cloud, reliability, and real-time data. He founded Payconiq (ING spinoff) and led ING Direct France IT, delivering early mobile banking innovations; earlier roles include LVMH and HP. Antoine holds a BSc from UTC (Compiegne – France) and an MSc from North Carolina State University. You can contact Antoine at larmanjat@google.com.

Philippe-Emmanuel DouziechCAST
Philippe-Emmanuel Douziech

Principal Research Scientist & Director of AI Research Programs

Philippe-Emmanuel Douziech leads AI programs at CAST Research Lab as Principal Research Scientist and Director of AI Programs. He focuses on advancing AI-driven software intelligence – turning research on code understanding, architecture mapping, and modernization into practical capabilities for customers and partners. He drives collaborations across industry and academia and contributes to applied research and publications. Philippe-Emmanuel holds a master’s degree from Mines Paris. You can contact Philippe-Emmanuel at p.douziech@castsoftware.com.

Understanding Large Custom Applications with Google AI & CAST

A joint CAST and Google Cloud paper on using generative artificial intelligence for code understanding at scale, and improving answer quality by augmenting it with CAST’s deterministic structural and dependency intelligence for large, multi-repository applications.

Clarify the limitation: generative artificial intelligence can explain a file well, but struggles with thousands of files and millions of lines of code without reliable structural context.

Describe an “agent” approach and show how adding CAST structure and context improves outcomes: impact analysis, risk detection, precise testing scope, transformation decisions, and better rationale.

Outline the setup: generate hierarchical documentation, store code and explanations in a vector database, and then run CAST Imaging structural analysis to build a graph-based “digital twin” of code elements and cross-technology flows.

Provide examples of the kinds of modernization questions addressed (identifying components, verifying real usage in call graphs, mapping program flows, and pinpointing affected interfaces and data entities).