Devproof-Ai

DevProof AI — Don’t just suggest a fix. Prove it works. AI coding assistants can quickly suggest code changes, but a suggested fix is not necessarily a correct fix. Developers still need to understand the root cause, review the change, write tests, execute them, and determine whether the reported issue was actually resolved. DevProof AI turns this process into a single verification-focused workflow. A developer selects a sample project or uploads Python source files and describes a bug. DevProof AI analyzes the project, identifies relevant files and the probable root cause, and proposes a code fix. The developer remains in control and must explicitly approve the proposed fix before verification begins. After approval, DevProof AI generates targeted pytest tests designed to validate the proposed behavior. The tests are executed inside a controlled project workspace, and the system produces a verification report containing the verdict, generated tests, test output, relevant files, root cause, fix explanation, and test counts. Our demo uses an order-service coupon calculation bug where the coupon is incorrectly applied before tax. DevProof AI identifies the problematic code, proposes the corrected calculation, generates tests that distinguish the buggy and corrected behavior, and executes them. The final result shows 3/3 tests passed, providing evidence that the proposed behavior works. The system is built with React, Vite, TypeScript, Tailwind CSS, Python, FastAPI, SQLite, REST APIs, and pytest. An LLM provider abstraction supports IBM watsonx, OpenAI, and a Mock provider for local development. IBM Bob 2.0 was used throughout development for project scaffolding, frontend and backend implementation, testing, debugging, security improvements, and workflow development. Bob task session summaries from both developers are included in the repository. DevProof AI focuses on a simple principle: AI can suggest a fix, but automated verification should provide the proof.