Design a complete LLMOps pipeline for a production AI product. Starting from raw data, walk through every stage: data collection and labeling, preprocessing and quality filtering, model training or fine-tuning, evaluation and red-teaming, staging and canary deployment, serving infrastructure, monitoring, and feedback loop back to training. For each stage, identify the key failure modes, the tools you would use, and the metrics you would track. What does the feedback loop look like — how does production behavior improve future model versions?