🚀 JDK Modernizer Agent

AI-powered autonomous migration from Legacy Java 17 to JDK 25.

JDK Modernizer is a stateful AI agent that migrates legacy Java codebases using a "Refactor-Verify-Correct" loop. It bridges the gap between technical debt and modern performance by combining the deterministic precision of OpenRewrite with the creative reasoning of Llama 3.3 (via Groq).

---

🚀 Problem Statement

Legacy Enterprise Java migrations are often stalled due to lack of time, effort, and priority. Manual refactoring is:

1. Error-Prone: Developers often miss deprecated API edge cases or complex switch exhaustiveness.
2. Scale-Limited: Large monorepos require consistent application of recipes across thousands of files.
3. Performance-Blind: Legacy concurrency models (Fixed Thread Pools) aren't automatically converted to Virtual Threads, leaving performance gains on the table.

🛠 Approach

We utilize a Hybrid Intelligence model:

• Deterministic Logic (OpenRewrite): Handles bulk, safe transformations using Lossless Semantic Trees (LST) to ensure 100% syntactic integrity for standard migrations (up to Java 21).
• Probabilistic Logic (LLM): Orchestrates the migration to JDK 25, identifies complex refactoring targets (like converting POJOs to Records), and self-corrects compilation errors.
• Stateful Orchestration (LangGraph): Manages the migration lifecycle, allowing the agent to "backtrack" or retry if a specific refactor breaks the build.

---

🏗 Architecture Diagram

graph TD
    A[Legacy Java 17 Code] --> B[Agent Scanner]
    B --> C{Agent Loop}
    
    subgraph Refactoring_Engine
        C --> D[OpenRewrite Node]
        D -- "Java 21 Recipes" --> E[Update Pom.xml / API Cleanup]
    end

    subgraph AI_Reasoning
        E --> F[LLM Java 25 Pass]
        F -- "JDK 25 Knowledge" --> G[Records & Pattern Matching]
    end

    G --> H[Verification Node: Maven]
    H -- "Fails" --> F
    H -- "Success" --> I[Modernized JDK 25 Code]

Loading

📦 Components

LangGraph Orchestrator: A cyclic graph that manages the state of the migration (Current File, Error Logs, Attempt Count).

OpenRewrite Worker: Executes standard JEP (Java Enhancement Proposal) recipes (e.g., UpgradeToJava25).

Maven Validator: A specialized node that triggers ./mvnw compile and parses the output to feed errors back to the LLM.

CopilotKit UI: A side-by-side Diff View that lets users review and approve AI-proposed changes before they are committed to disk.

🔄 OpenRewrite Integration

The project leverages the rewrite-migrate-java artifact. We programmatically trigger recipes like:

org.openrewrite.java.migrate.UpgradeToJava25

org.openrewrite.java.migrate.net.MigrateURLConstructorToURI (handling JDK 20+ deprecations)

🧠 LLM & Inference Strategy The agent currently utilizes Groq Cloud for high-speed, low-latency inference.

• Primary Model: llama-3.3-70b-versatile (via Groq API).
• Reasoning: Chosen for its superior performance in Java code generation and zero-cost free tier for developers.
• TODO: - [ ] Add ChatOllama support to allow local execution using DeepSeek-Coder-V2.

🏗 Build (Maven)

The agent interacts directly with the Maven lifecycle.

Phase 1: Update <java.version> and compiler plugins in the pom.xml.

Phase 2: Execute a "Dry Run" of OpenRewrite.

Phase 3: Iterative compilation. If the build fails, the agent isolates the error and re-attempts the refactor on the specific file.