Meta Data Engineering Onsite Interview Experience and Insights

There are many interview experiences shared online, and I've consulted quite a few. I'm sharing my own interview experience after my own, hoping it will be helpful to everyone. Best of luck in getting

There are many interview experiences shared online, and I've consulted quite a few. I'm sharing my own interview experience after my own, hoping it will be helpful to everyone. Best of luck in getting offers! I had three interview questions: DoorDash, Netflix, and one old question. Here's a list of what I remember: DoorDash: The Product Sense question mainly revolved around restaurants, such as: what metrics to look at for launched restaurants, what dimensions to use to segment the data, how to visualize it, and possible reasons for declining restaurant revenue. These common questions are all covered in other interview experiences, so I won't go into detail. Here, I want to emphasize how to improve user loyalty—note that this isn't about engagement or volume, but loyalty. Modeling was a classic orders model, containing two tables: orders and order_items. It was relatively simple, but there were some follow-ups, such as how to add a metric to measure market share, or how to handle a restaurant with multiple menus. The SQL question wasn't difficult; it mainly tested the orders and restaurant tables, such as calculating the pickup to delivery ratio and identifying the top 3 restaurant types. Python is relatively more difficult. The input is roughly like this: ordersteps: [{driver, action, location_id}, {driver, action, order_id}] When action=travel, there is no order_id; When action=pickup/dropoff, there is no location_id. There is also a timetable: time: {(0, 1): 10} Required output: order xx delivered in xx mins Netflix: Product sense leans towards content & engagement, for example: How to measure the success of a newly launched show? Possible drivers behind user binge-watching behavior. How to investigate if the click-through rate of the recommendation system drops? Modeling leans towards the relationship between users and content, usually involving tables of user viewing behavior (watch events) and program content (shows/movies).

Follow-ups may include: How to define and calculate “active user”? How to model a show with multiple seasons/episodes? How to design metrics to measure content diversity. SQL is quite practical, for example: calculating the DAU for a given day. Ranking viewership duration by content type (movies, TV series, documentaries). Identifying churn users: subscribers who haven't watched anything in the past 30 days. Python is more suited for data processing and algorithm problems, such as: given a user's viewing time series, calculating their longest consecutive binge session. Simulating a recommendation logic, such as generating recommendation candidates given user profiles and program tags.