MonoSplitter AI

Problem: Breaking a monolith into microservices is one of the most common and costly application-maintenance tasks in enterprise software. Code, imports, and database models get tightly coupled across dozens or hundreds of files, and manually tracing every coupling point, rewriting call sites, and verifying nothing broke can take a developer weeks to months — with high risk of missed dependencies and production incidents. Solution — MonoSplitter AI: A tool that visualizes a monolith's module and database coupling as an interactive graph, then uses IBM Bob to agentically carve a selected module out into a standalone microservice — end to end, not just code suggestions. The dashboard runs madge against a real Node/Express monolith to generate a genuine code-dependency graph (not simulated), overlays it with the database schema's foreign-key relationships, and surfaces every coupling point in plain language — e.g., "users/index.js calls processPayment() directly, in-process" — so a developer sees exactly where the risk is before touching anything. From there, a single "Carve Out with Bob" action triggers Bob to: copy the target module into its own service, scaffold a standalone Express app and Dockerfile, rewrite the in-process import into an HTTP client call, and generate a contract test that verifies the new service behaves identically to the old code path — both happy-path and error cases. Bob then goes further, generating the split database schema, an isolated data-access layer, and a full production migration plan (dual-write → backfill → read cutover → write cutover → cleanup) as documentation, without touching a real database. Target users: engineering teams maintaining monoliths who need to extract services incrementally and safely — the exact audience the challenge's "application maintenance" workflow describes. Why it's creative and unique: most AI coding tools assist with writing code one file at a time. This tool orchestrates a full workflow — visualization, decision-making, multi-file agentic execution, and verification — and proves its own correctness: the contract test isn't decorative, it actually runs against both the old and new code paths and the demo only claims success once it genuinely passes. Every number shown in the UI (files scanned, coupling points found) traces back to real tool output, not hardcoded demo data. Effectiveness: the tool collapses a task that traditionally takes a developer hours to weeks of manual tracing and rewriting into a single reviewable action, with automated proof of correctness built in rather than left to the developer to write afterward.