Focuses on writing maintainable, readable, and robust code together with practical debugging approaches. Candidates should demonstrate principles of clean code such as meaningful naming, clear function and module boundaries, avoidance of magic numbers, single responsibility and separation of concerns, and sensible organization and commenting. Include practices for catching and preventing bugs: mental and unit testing of edge cases, assertions and input validation, structured error handling, logging for observability, and use of static analysis and linters. Describe debugging workflows for finding and fixing defects in your own code including reproducing failures, minimizing test cases, bisecting changes, using tests and instrumentation, and collaborating with peers through code reviews and pair debugging. Emphasize refactoring, test driven development, and continuous improvements that reduce defect surface and make future debugging easier.
MediumTechnical
0 practiced
After a refactor you deploy a new model server and observe increased inference latency for the same model and hardware. Describe a methodical debugging and rollback plan: how to profile, compare baselines, identify culprit changes, run canary deployments, and roll back safely while minimizing user impact.
MediumTechnical
0 practiced
Implement a pytest fixture named 'deterministic_env' that sets seeds for Python random, NumPy, and PyTorch (CPU and CUDA), sets torch.backends.cudnn.deterministic and benchmark flags appropriately, and yields control to tests before restoring previous state. Provide the fixture code and an example test that uses it.
HardSystem Design
0 practiced
Design an observability dashboard and alerting policy to detect production model performance drift and degradation. Specify which metrics (e.g., latency, throughput, prediction distribution shifts, AUC on sampled labeled data, input feature statistics), aggregation windows, thresholds, and alert actions (paged, ticketed, auto-rollback).
EasyTechnical
0 practiced
You are reviewing a repository where data preprocessing, model architecture, training loop, and evaluation are all implemented in one 2000-line Python file. Explain the Single Responsibility Principle (SRP) and provide concrete, actionable refactoring steps to split this codebase into testable modules (data, models, training, evaluation, utils). Describe what public APIs each module should expose and how you'd structure tests and CI to enforce the separation of concerns.
MediumTechnical
0 practiced
Write a set of pytest parametrized tests for a tokenization function 'tokenize(text)' that cover edge cases: empty string, multiple consecutive whitespaces, non-ASCII characters (emoji), very long input, and mixed punctuation. Use pytest.mark.parametrize and describe expected tokens for each case.
Unlock Full Question Bank
Get access to hundreds of Code Quality and Debugging Practices interview questions and detailed answers.
