Interview Practice Questions

Design an architecture for a CI/CD process in a machine learning project that includes 10 developers and 5 models operating in real-time. How will you ensure automation of testing and deployment of models? What DevOps principles will you apply?

Design an active learning system for a model that needs to process 1 terabyte of data per day with a requirement to label 10% of the data. How will you organize the labeling process considering you have a limited budget and a team of 5 annotators?

You need to build a deployment system for a machine learning model that will handle 2000 requests per second with a latency of no more than 200 milliseconds. How will you approach the choice between batch and real-time deployment? What architectural decisions will you propose?

Design a system for storing and processing metadata about machine learning models. Given that you will have 50 different models and 1000 metadata entries for each model, how will you ensure the availability and performance of the system? What database would you use?

You need to build a system that tracks changes in the data used for training models. Given that there will be 5 different data sources in the system, how will you organize the monitoring and alerting process for data drift? What technologies will you use for analysis and visualization?

Design a system that monitors and alerts for data drift and concept drift in a machine learning model. You have a budget constraint of $100,000 per year, and the system should handle 10 million records per month. The response time for alerts should not exceed 5 minutes.

How would you design a video recommendation system for an e-commerce platform?

How would you deploy a recommendation model to production?

How would you design a fraud detection system?

How would you design fraud detection for banking transactions?

How is an ML model deployed to production?

How would you encode a product with 7,000 heterogeneous attributes for a neural recommendation model?

How would you build a user embedding from the user's sequence of actions?

How would you combine user and product embeddings to predict relevance?

How would you optimize a computationally expensive recommendation model for production inference?

How would you build a system that generates product listing descriptions when launching in a new country, China?

How would you design an offline metric for text generation quality?

Describe the complete ML pipeline from data collection to inference: what stages and tools did you use?

Do you have experience optimizing inference with TensorRT, quantization, or similar approaches?

What tools do you use for experiment tracking and model versioning?