AI-Powered Architecture Discovery for a Legacy Monolith

An enterprise-grade policy administration and claims platform serving multiple lines (auto, property, liability, etc.) needed a revamp to improve performance, reduce operating expenses, and boost user experience.

Since major modules included quotes, underwriting, endorsements, billing, collections, loss notices, indemnification, and subrogation, we needed to map dependencies within the app and workflow to ensure the smooth, secure transition.

The app core is represented by a tightly coupled monolith with embedded ratings, rules, document generation, and batch jobs for nightly bordereaux, commissions, and compliance reporting. Also, there were connections to payment gateways, credit bureaus, third-party data providers, and regulatory e-filing.

Recommended by industry experts, the client sought to eliminate operational pain points, opaque dependencies, manual handoffs, and fragile CI/CD. Devox Software’s team got to work, making a zero-downtime, evidence-driven modernization kit.

The team roles were flexibly distributed according to the project phase and involved a Delivery Manager, a Solution Architect, a Senior Backend Engineer, a Data and ML Engineer, a DevOps Engineer, a QA Engineer, and a Business Analyst with experience in InsurTech.

The project showed the most common difficulties when trying to upgrade a legacy app:

• State-by-State Regulatory Variability. Different states enforced different reporting requirements, policy rules, audit standards, and compliance procedures. This complicated modernization because even a small re-architecture could affect downstream compliance logic.
• Actuarial Dependency Risks. Rating engines, underwriting logic, and scoring models required actuarial validation and business approval before changes could be introduced into production.
• External System Coupling. Critical workflows depended on external reinsurers, regulatory services, payment providers, and reporting systems. Many integrations lacked centralized ownership or clear operational visibility. 
• Technical Limitations. Sparse documentation, tight coupling across policy, billing, claims, and reporting services, brittle CI/CD; inconsistent data contracts; mixed tech generations, and more.

Instead of starting with manual reverse engineering, Devox Software designed an AI-assisted architecture discovery pipeline that combined static analysis, runtime telemetry, and LLM-powered semantic interpretation.

Phase 1. Static Code & Dependency Analysis

We first ingested source code repositories, SQL scripts, CI/CD configurations, infrastructure manifests, and deployment pipelines.

The system then analyzed call graphs, database relationships, service interactions, event flows, and shared dependency chains.

This revealed hidden coupling between policy, claims, billing, and reporting domains.

Phase 2. LLM-Assisted Semantic Understanding

To understand undocumented business logic, we applied LLM-assisted semantic analysis across legacy service methods, stored procedures, batch jobs, and integration adapters to classify domain responsibilities, identify duplicated logic, detect anti-patterns, and generate contextual architectural documentation.

This significantly accelerated dependency discovery compared to manual analysis alone.

Phase 3. Runtime Correlation with Observability Data

Static analysis alone was insufficient because many real production flows differed from repository assumptions. To solve this, we correlated legacy architecture data with OpenTelemetry traces, distributed transaction telemetry, infrastructure metrics, and production traffic patterns visualized in Grafana.

This allowed us to identify actual runtime bottlenecks and critical user journeys affecting claims processing.

Modernization Strategy

Based on the AI-generated architecture map, Devox Software designed a phased strangler-fig modernization roadmap.The approach focused on extracting low-risk domains first and gradually shifting workloads into modernized services. This way, the modernization pipeline included:

• canary releases
• automated rollback
• pre-flight validation
• infrastructure-as-code standardization
• and progressive traffic shifting inside Azure AKS

This minimized operational risk while preserving business continuity.

“One of the most important engineering decisions was choosing incremental modernization over a full rewrite. While a complete rebuild could theoretically simplify architecture faster, it introduced unacceptable operational and regulatory risk,” said Alex Kukarenko, Director of Legacy Systems Modernization at Devox Software.

We’ve created a comprehensive app architecture mapping to prepare for further modernization steps. This way, the client has received the following outcomes:

• Accelerated dependency discovery and migration planning by approximately 45%
• Reduced Sev-1 incidents during deployments by nearly 60%
• Improved incident investigation speed by up to 70% through OpenTelemetry distributed tracing
• Enabled rollback validation and canary releases with under 5-minute rollback readiness
• Reduced manual dependency analysis and architecture reverse-engineering effort by approximately 50%
• Built a domain-oriented modernization roadmap supporting phased migration to AKS and event-driven services

As a result, the following modernization has been carried out without downtime.