Top Python Interview Questions

Python interviews split sharply by role, and preparing for the wrong split is the most common wasted effort. Backend interviews probe the data model — mutability, scoping, decorators, generators, the GIL. Data and ML interviews probe pandas, NumPy semantics and statistics, and treat the language as a given. Automation and SRE interviews probe the standard library and process handling.

What is common to all three is that Python's convenience hides mechanism, and interviewers probe exactly there. Mutable default arguments, late-binding closures and identity versus equality are asked so often because they are where the surface simplicity stops matching what the interpreter does.

The questions that separate users from understanders

# Evaluated once, at definition - the classic
def append(item, bucket=[]):
    bucket.append(item)
    return bucket
append(1); append(2)          # [1, 2] - one shared list

def append(item, bucket=None):
    bucket = [] if bucket is None else bucket
    return bucket + [item]

# Late binding: the closure captures the variable, not its value
fns = [lambda: i for i in range(3)]
[f() for f in fns]            # [2, 2, 2]
fns = [lambda i=i: i for i in range(3)]
[f() for f in fns]            # [0, 1, 2]

Concurrency: pick the right tool out loud

The single most useful thing you can demonstrate is knowing which of Python's three concurrency models fits a given workload, and why. Say the classification before the solution and most follow-up questions answer themselves.

I/O-bound, many operations: asyncio — thousands of outstanding operations on one thread, no GIL contention because nothing is computing. I/O-bound, blocking libraries: threads, because the GIL is released during I/O. CPU-bound: multiprocessing or a native extension that releases the GIL — NumPy does this, which is why array operations scale where pure-Python loops do not.

Data roles: pandas semantics get asked more than algorithms

For data and ML positions, the language questions thin out and are replaced by questions about pandas behaviour — which is reasonable, because that is where the bugs are. The SettingWithCopyWarning, the difference between loc and iloc, why chained indexing is unreliable, how merges duplicate rows, and how NaN propagates through aggregations all come up repeatedly.

The underlying theme matches the database section of any backend interview: know whether you are looking at a view or a copy, and know what a join does to your row count. Those two questions catch most real pandas bugs.