Hey HN, OP here. I've spent the last few months building a tool to help migrate legacy Oracle systems, specifically the heavy business logic trapped in PL/SQL packages. The Problem: Most migration tools fail on complex logic like CONNECT BY (recursion), SELECT FOR UPDATE (locking), or extensive cursor loops. LLMs alone hallucinate variables or forget context. The Solution: I built a pipeline that mixes AST parsing with LLMs (Claude Sonnet/Gemini). It works in stages: Pre-Analysis: Maps the schema and detects complexity patterns. Planning: Decides the architecture (Services, DTOs). Generation: Converts SQL to Java 17/Spring Boot 3.3. Self-Healing: A feedback loop that tries to compile the code in a Docker sandbox and fixes syntax errors automatically before delivering. It handles recursive queries by converting them to WITH RECURSIVE CTEs in JPQL/Native Query, and maps FOR UPDATE to JPA Pessimistic Locks. It's running on Node.js with BullMQ/Redis for the async processing queue. I'd love to hear your feedback on the generated code quality. Link to demo: https://blog.thalesouza.com.br/en/blog/logic-migration/