Ds Python Interview — Data skill for Claude Code

ds-python-interview

A [Claude Code](https://claude.com/claude-code) skill for practicing **Python data-science interview questions** in Jupyter notebooks, with built-in **spaced repetition**.

It splits the work cleanly:

• A standard-library-only CLI does the bookkeeping — storing questions, generating notebooks, tracking what's due, and updating the spaced-repetition schedule.
• Claude does the reasoning — writing grounded questions, parsing pasted text or screenshots, producing model solutions, and reviewing your completed notebooks against an answer key.

Questions span four categories:

• dsa — pure-Python / standard library (collections, itertools, heapq, sliding window, two pointers, DP, clean idiomatic code).
• pandas — data manipulation (groupby/transform/merge/reshape, window functions, vectorization; numpy broadcasting; scipy.stats).
• stats — implement an estimator or test (CUPED, delta method, IPTW, bootstrap, power/MDE) from scratch or with statsmodels/scipy.
• sql — analytics SQL answered as real queries inside the notebook: the setup cell loads small DataFrames into an in-memory database — DuckDB (Postgres-style dialect) if installed, else stdlib SQLite with a DATE_TRUNC shim — and defines a q() helper, so joins, window functions, and retention/NDR-style metrics run with zero required dependencies. Supports multi-step interview cases (one business scenario ramped easy → hard across Q1→QN, with a planted data trap) — ideal for giving or receiving a mock SQL interview.

ML-implemented-from-scratch is intentionally out of scope.

What it does

1. Generate a notebook — pick a category + difficulty; get a working notebook with sequential questions Q1 → QN. The intro cell lists the packages required to run the notebook (auto-derived from the questions' code). Each question shows the input data and expected output as a rendered table (a markdo