Банк вопросов из реальных собеседований

You can run four Python applications as systemd services on one VM or as four containers. What practical guarantees do containers add?

How are matrix equation, least squares, gradient descent and L1/L2 regularization related?

Marketing wants to know how much it can pay for acquisition without losing money. Why might MSE be insufficient for LTV prediction, and what metric or objective would you add?

What exactly should be counted in metrics: timestamps, operation duration, or error values themselves?

For a new player, you need to understand when you can already trust the VIP status forecast. How to assess at what day in a client’s life the model gives a sufficiently useful signal?

For a new player, you need to understand when you can already trust the VIP status forecast. How to assess at what day in a client’s life the model gives a sufficiently useful signal?

You review code that loops over texts, calls an embedding model one by one and appends outputs to a NumPy array. What would you improve?

What does it mean to take an ML model from training to production, and which pieces should an ML engineer be able to own?

You have subscription transaction history and need to estimate or predict yearly LTV for a cohort. How do you define the target and build a first baseline without leaking the future?

The backend service on Python became slow under load. How would you localize the bottleneck and understand what exactly to optimize?

How can you avoid blocking the system if resource cleanup or fallback script may also fail?

When to choose threads, when individual processes, and how does GIL affect I/O and computation tasks?

When to choose threads, when individual processes, and how does GIL affect I/O and computation tasks?

What is the difference between flows and processes? What is GIL in CPython when synchronization and IPC are needed?

Explain how the hash table in the Python dict is structured and why lookup/insert operations are usually O(1) but sometimes degrade.

How to check the student's code and give a hint without revealing the finished solution?

You need to forecast how long a ship will wait at a port. How would you define the ML target, time granularity and prediction horizon so the result is useful for operations?

Describe how a model moves from notebook research to production training, versioned model storage and serving for an LTV prediction service.

Describe how a model moves from notebook research to production training, versioned model storage and serving for an LTV prediction service.

How much does it take to insert n different strings of length k into a Python set? What happens when hash returns the same value for all objects?

What is the difficulty of adding an item to the beginning and end of the Python list? Why does an append run on average for O(1), but a separate call may require O(n)?

How to deconstruct Python code with lists, object references, cyclic references, and variable default argument values and explain what will remain in memory?

How to count attempts in the retry decorator so that the logs and metrics reflect the real number of calls?

What design is needed for a Python service that processes match events, builds features, and runs an ML model for real-time analytics?