Design a response-ranking ML system

Design an end-to-end machine learning system that ranks multiple candidate text responses for a user query to maximize user satisfaction. Specify: problem formulation and objective (labels or proxies), data sources and labeling strategy (implicit feedback, human ratings), model choice (e.g., pairwise or listwise ranking, or RL from feedback), offline training pipeline and embedding/feature generation, evaluation metrics (e.g., NDCG, pairwise accuracy, calibration), online inference architecture (latency budget, caching, candidate generation), experimentation plan (A/B testing, counterfactual evaluation), safety and alignment measures (toxicity filters, guardrails), bias/privacy controls, monitoring and alerting, retraining cadence, and cost/reliability trade-offs. Provide a high-level architecture description in words.