roadmap
- 01
- 02
- 03
- 04
- 05
- 06
- 07
- 08
Explain Python at the depth interviews probe, from the data model to how the interpreter runs your code. Covers core types, functions and classes, typing, concurrency and asyncio, packaging and testing.
1717 of 1717 questions in your plan: 0 known, 0 in progress, 0 due, 0 skipped, 1717 new. 0 cut, outside your plan.
Few postings ask for Python and nothing else. The language usually arrives inside another role, and that role decides where the questions go once the warm-up on names and objects is over. Backend and platform teams push furthest into the runtime: a blocking call inside a coroutine that stalls every request on the event loop, threads that make CPU-bound work no faster, a worker whose cleanup never runs when its container sends SIGTERM. Data and ML loops, for the [Data Engineer](/roadmaps/data-engineer) and [Machine Learning Engineer](/roadmaps/machine-learning) roles among others, stay nearer the object model and iteration: a generator that streams a file too large for memory, a shallow copy of nested lists whose rows are still shared, a lambda that a `multiprocessing` pool cannot pickle. [MLOps](/roadmaps/mlops) interviews add environments and pinned dependencies, since a model is only as reproducible as the environment that trained it. [AI Engineer](/roadmaps/ai-engineer) loops sit between the two camps, with concurrent calls to model APIs and typed shapes for what comes back. Security roles read Python as much as they write it. [DevSecOps](/roadmaps/devsecops) and [Cyber Security Expert](/roadmaps/cyber-security) loops ask why `pickle.loads` on a downloaded file runs someone else's code, what `shell=True` does to a filename with a semicolon in it, and why a reset token comes from `secrets` and not `random`. Each of those role roadmaps sets this tree beside the other subjects its loop covers.
The [Python hub](/topics/lang-python) carries this tree's primer, the map of which sections lean on which; drilling the plan's questions shows which parts of that map you can defend out loud. Return to the primer when several misses trace back to one idea. A junior is judged first on [mutability, identity and copying](/topics/lang-python-core-types-mutability-identity): what `b = a` actually binds, when `is` and `==` disagree, why a list default argument remembers earlier calls. Get that solid before anything else, because most later answers lean on it without saying so. The step to middle level usually runs through [generator functions](/topics/lang-python-decorators-generators-generators) and the iterator protocol beneath them: lazy pipelines, `yield from`, a generator that is empty on its second pass. Senior candidates reach [concurrency](/topics/lang-python-concurrency) early, because the question there is a judgment rather than a definition: threads, processes or asyncio for an I/O-bound crawler, a CPU-bound transform, a service holding thousands of open sockets, and what the GIL, or a free-threaded build, changes in each case. Fluency in another language is a reason to narrow importance to must know and let the default study order, by frequency, decide what comes up: the questions asked most tend to sit where Python breaks habits brought from elsewhere. Someone learning the language through data work should switch to by topics instead, so the language core arrives before everything that assumes it.
Few postings ask for Python and nothing else. The language usually arrives inside another role, and that role decides where the questions go once the warm-up on names and objects is over. Backend and platform teams push furthest into the runtime: a blocking call inside a coroutine that stalls every request on the event loop, threads that make CPU-bound work no faster, a worker whose cleanup never runs when its container sends SIGTERM.
Data and ML loops, for the Data Engineer and Machine Learning Engineer roles among others, stay nearer the object model and iteration: a generator that streams a file too large for memory, a shallow copy of nested lists whose rows are still shared, a lambda that a multiprocessing pool cannot pickle. MLOps interviews add environments and pinned dependencies, since a model is only as reproducible as the environment that trained it. AI Engineer loops sit between the two camps, with concurrent calls to model APIs and typed shapes for what comes back.
Security roles read Python as much as they write it. DevSecOps and Cyber Security Expert loops ask why pickle.loads on a downloaded file runs someone else's code, what shell=True does to a filename with a semicolon in it, and why a reset token comes from secrets and not random. Each of those role roadmaps sets this tree beside the other subjects its loop covers.
The Python hub carries this tree's primer, the map of which sections lean on which; drilling the plan's questions shows which parts of that map you can defend out loud. Return to the primer when several misses trace back to one idea.
A junior is judged first on mutability, identity and copying: what b = a actually binds, when is and == disagree, why a list default argument remembers earlier calls. Get that solid before anything else, because most later answers lean on it without saying so. The step to middle level usually runs through generator functions and the iterator protocol beneath them: lazy pipelines, yield from, a generator that is empty on its second pass. Senior candidates reach concurrency early, because the question there is a judgment rather than a definition: threads, processes or asyncio for an I/O-bound crawler, a CPU-bound transform, a service holding thousands of open sockets, and what the GIL, or a free-threaded build, changes in each case.
Fluency in another language is a reason to narrow importance to must know and let the default study order, by frequency, decide what comes up: the questions asked most tend to sit where Python breaks habits brought from elsewhere. Someone learning the language through data work should switch to by topics instead, so the language core arrives before everything that assumes it.