Whenever you're ready, walk me through how you'd approach this.
Given a dataset with customer purchase history, how would you approach building a recommendation system at Amazon Web Services?
Discuss how you would represent customer and item interactions, handle implicit feedback and sparse data, generate candidates, select and train a ranking model, and prevent temporal leakage. Explain how you would evaluate recommendations offline and validate them with an online experiment. Address cold-start users and items, latency, scalability, personalization, and AWS deployment choices. Provide production-quality Python demonstrating preprocessing, training, and Recall@K and NDCG@K evaluation.