Python Interview Preparation with AI: What to Practise and How

“Python developer” is three different jobs wearing one job title. A backend interview asks about frameworks, databases, and concurrency. A data interview asks about pandas, SQL, and statistics. An ML interview asks about models, metrics, and pipelines. Preparing for the wrong one is the most common way strong Python candidates fail. Start by deciding which loop you are actually entering.

The core Python everyone gets asked

Whatever the track, expect a shared base: mutable versus immutable types and the classic mutable-default-argument trap, list comprehensions versus generators and why generators matter for memory, decorators (be ready to write one), context managers and what with guarantees, and *args/**kwargs. Then the two questions that separate levels: how does Python manage memory (reference counting plus a cycle collector), and what is the GIL.

Answer the GIL question by its consequence, not its definition: threads do not give you parallel CPU work, so use multiprocessing or native extensions for CPU-bound work and threads or asyncio for I/O-bound work. Mentioning that recent Python versions are moving toward an optional free-threaded build shows you are current.

Track one: backend

Django or FastAPI, and why you would pick one over the other. ORM questions almost always include the N+1 query problem — name it, then say select_related / prefetch_related or the equivalent join, and how you would have caught it with query logging. Expect asyncio: what an event loop is, why a blocking call inside an async handler poisons the whole loop, and where you would put a slow task instead. Then REST design, authentication, caching, and one system design question.

Track two: data

SQL is usually the real filter, not Python: joins, window functions, group-by with filtering, and the ability to say what a query returns before running it. On the Python side, pandas transformations, merges, missing data, and why you avoid iterating over rows. Statistics questions cluster around A/B tests, p-values, sampling bias, and how you would explain a result to a product manager who does not care about your method.

Track three: ML

Overfitting and how you detect it, train/validation/test discipline and leakage, why accuracy is the wrong metric for imbalanced data, feature engineering choices, and how a model gets to production and stays healthy there. The strongest signal you can give is one project you can defend end to end: the problem, the data, what you tried, what failed, and what the metric was at the end.

The coding round

Python coding rounds are string, dictionary, and set problems more often than graph theory. Use the standard library — collections.Counter, defaultdict, itertools — because fluency with it is part of what is being scored. State the complexity before you write, and narrate as you go.

How to practise with an AI interviewer

Pick the track, set the experience level, and answer out loud in the language your real interview will use. Do the GIL explanation and the N+1 explanation until each takes about a minute cleanly. If you are on the data or ML track, rehearse defending your project as a spoken story, because that is what the interviewer will ask for and it is the answer people least often practise. Repeat sessions until the feedback stops repeating itself.