CloudBlueprintAI: An Intelligent Framework for Reverse Engineering Cloud Architecture from Source Code and Infrastructure-as-Code

Abstract

Cloud-native enterprise systems increasingly rely on distributed infrastructure managed through Infrastructure-as-Code (IaC) artifacts, CI/CD pipelines, and containerisation platforms. As infrastructure evolves continuously through automated deployment pipelines, architecture documentation becomes decoupled from the actual deployed state, creating what practitioners term architecture drift. This drift imposes substantial costs during security audits, cloud migrations, and modernisation initiatives, as engineering teams must invest significant manual effort to reconstruct an accurate picture of deployed systems. Existing architecture recovery approaches focus primarily on software component dependencies and fail to reconstruct the complete cloud infrastructure layer—including network topology, deployment paths, resource dependency graphs, and governance configurations.This paper introduces CloudBlueprintAI, a five-component intelligent framework designed to reverse engineer cloud architecture automatically from multi-artifact software repositories. The framework comprises: the Unified Repository Intelligence Engine (URIE), which ingests and analyses source code, Terraform and Bicep templates, ARM templates, Kubernetes manifests, Helm charts, and CI/CD pipeline definitions in a unified semantic model; the Cloud Resource Inference Algorithm (CRIA), which maps SDK usage patterns and IaC declarations to deployed cloud resources; the Architecture Knowledge Graph (AKG), which encodes reconstructed architecture as a queryable semantic graph; the Intelligent Architecture Reconstruction Engine (IARE), which produces multi-dimensional architecture views from the knowledge graph; and the Architecture Confidence Scoring Algorithm (ACSA), which assigns per-resource confidence scores to guide manual validation. Evaluated against enterprise Azure repositories, CloudBlueprintAI demonstrates substantially higher reconstruction completeness compared to source-code-only baselines, with ACSA confidence scores ranging from 76% for network-layer resources to 98% for storage-layer resources. The framework reduces architecture documentation effort while enabling automated impact analysis, governance queries, and dependency reasoning.