Python notebooks right in the browser — nothing to install. Paired with pandas it's your main tool for cleaning data, EDA and your first API calls (Level 2).

Write SQL against big real-world datasets right in the browser, with hundreds of ready-made example queries next to them to learn from. Perfect for drilling SQL before you install anything locally (Level 1).

A friendly BI tool: connect a database and build dashboards with almost no code. Being able to present metrics cleanly stands out in a portfolio (Level 3+).

Every project is a repo with a README (question → data → what you found). In tech this is your living résumé — employers look here first, so start one from day one, even if it's messy.

A rock-solid, all-purpose relational database. Not the fastest for heavy analytics, but simple, familiar and more than enough to start — a great home for your dbt marts in a pet project.

A blazing-fast columnar database for analytics at scale and in real time: it chews through billions of rows and loves event data. The workhorse of serious product analytics.

A compressed columnar format — the de facto standard for analytical data: far smaller than CSV and faster to read because it loads only the columns you need. Rule of thumb: store data as Parquet between pipeline steps, not CSV.

The most widely used orchestrator: pipelines are Python DAGs, with a huge ecosystem of integrations and the most job-market demand. Heavier to set up — but if you learn one grown-up orchestrator, make it this one (Level 4).

A lighter, more Pythonic orchestrator: plain functions become tasks and flows via @task/@flow decorators, with retries and caching built in. Reach for it on your first pipeline to feel orchestration without the pain (Level 4).

The base kit of tests and distributions: t-test, chi-square, correlations, plus the normal and other distributions. The first thing you import once you get to hypothesis testing (Level 3).

Regressions and models with grown-up output: coefficients, p-values, confidence intervals, R². What you want for relationships, regression and any serious analysis (Level 4).

A high-level library: one-line tests that already return the effect size and confidence interval — the bits scipy makes you compute separately. Beginner-friendly.

Distributions and relationships visualized in one line: histograms, boxplots, scatter with a regression line. Plus built-in datasets (penguins, tips, titanic) that load instantly — perfect for the 'draw the data first' rule.

A free point-and-click app: t-tests, ANOVA, regression, correlations and even Bayesian versions — without a single line of code. Great for grasping the method itself without fighting syntax.

A JASP-style alternative: a friendly interface for the core tests and models, and a comfy bridge from Excel to real statistics. If one doesn't click, try the other — they're similar.

The simplest vector database for your first semantic search: drop in texts and query by meaning, not exact words. When you reach the AI layer, start here.

Runs language models locally on your own machine with one command — no cloud, no keys, no per-token bill. Handy for experimenting with models privately while you bolt an AI layer onto your analytics.

Wire up a language model over an API and build a tool on top: summarization, classification, 'a question in plain language → SQL over your dataset → an answer from the data'. That's basically a product.