You are building a personalized recommendation system for a consumer product. The team wants a plan for two cases: launching a new model with little live feedback, and serving new users who have little or no history.
How would you handle cold start for a new model or new users?