Whenever you're ready, walk me through how you'd approach this.
How would you fine-tune an open-source LLM for a banking chatbot using proprietary data at Capital One?
Explain the practical end-to-end approach, including data preparation, privacy controls, parameter-efficient fine-tuning, evaluation, deployment, monitoring, and rollback. Address how you would prevent sensitive-data leakage, preserve general capabilities, handle unsupported banking questions, and distinguish fine-tuning from retrieval-augmented generation. Provide production-quality Python code for a representative training pipeline and state the metrics and acceptance criteria you would establish before deployment.