Constitutional AI Critique Generator

Build an open-source implementation of Constitutional AI's critique and revision pipeline. **Goal:** Given a potentially harmful model output, generate: 1. A critique identifying specific problems 2. A revised output that addresses the issues 3. Explanation of changes made **Use Cases:** - Training data generation for safety fine-tuning - Real-time output filtering - Red-teaming feedback loops