1. Run embeddings against vectorized issues 2. Find most similar 3. Use LLM to compare Suggest the better one