I Built a RAG System That Knows When NOT to Answer — Here's How

Most AI projects I see are about making the model answer better. Mine is about making it refuse better. Here's the thing — in enterprise environments, a confident wrong answer is worse than no answer