Don't hunt for the "right" dataset — take the one you actually care about. Here are aggregators that hold anything, plus picks by theme. Each has what it is and what to ask it. Start with any of them.
Aggregators and search engines for any topic. Bookmark a couple — useful at any level.
The biggest hub: hundreds of thousands of datasets + competition data. Each has discussions, example notebooks and a free online coding environment.
The main AI hub: thousands of datasets (text, images, audio) next to models. Everything loads in one line of Python with streaming and caching.
Google tuned for datasets: searches the whole web — government portals, research repos, Kaggle. Know the topic but not where the data is? Start here.
A weekly newsletter of curious datasets: 5 a week, 400+ issues since 2015. A searchable archive — a trove of offbeat topics.
A huge categorized list of open datasets by topic (biology, finance, sport, transport…). All in one place, with links.
A registry of large open datasets in Amazon’s cloud: satellites, genomics, weather, transport. Often terabytes — for when you want to feel real scale.
Public datasets you query with SQL right in BigQuery: Wikipedia, weather, genomics, city data. Free tier: 1 TB of queries per month.
One of the oldest repos: clean, tidy tabular datasets for the classics (classification, regression). Perfect when you want something small and clear.
If you game — start here. The data is familiar and questions write themselves.
A fresh dump: 65,000+ Steam games (2021–2025) with genres, prices, tags. Great for your first aggregates and groupings.
6.4M player reviews labelled “recommend / not” — a ready playground for text analytics and NLP.
290 games with descriptions, genres, ratings and nearly a million reviews — handy for linking “game traits” to “player reaction”.
Mini recommender systems and taste analysis — very tangible.
The recommender classic: film ratings + genres and tags, from 100k to 32M rows. Perfect for your first JOINs — start with the small version.
IMDb’s official non-commercial dumps: films, series, ratings, actors and roles. Lots of data — good practice for cleaning and multi-table joins.
114,000 tracks with audio features: danceability, energy, tempo, “mood” (valence), popularity. Also mirrored on Hugging Face.
The most hireable. Data behaves like a real product: orders, users, funnels, churn.
Real (anonymized) data of a Brazilian marketplace, 2016–2018: 100,000 orders across 9 linked tables — orders, customers, products, sellers, payments, reviews, geolocation. Essentially a mini-model of a real product — this is what product SQL is actually taught on.
A compact churn classic: plans, services, tenure, left/stayed. Small and clear — good for segmentation and your first predictions.
Real transactions from a UK online shop: receipts, products, quantities, countries. A classic for RFM analysis and cohorts.
Event-stream generators like a streaming service (logins, plays, purchases): EventSim, Lenses Datagen, Mockingbird. You generate the data yourself — ideal for product metrics.
Huge datasets that let you honestly say "I've worked with billions of rows." Some run right in your browser — nothing to install.
Write SQL right in your browser against 35+ real datasets — Reddit and Hacker News posts, GitHub events, NOAA weather, forex. Each one ships with 220+ ready-made queries, so you can learn by taking apart someone else's SQL without installing a thing.
Every public GitHub action, logged hour by hour: commits, stars, PRs, forks. Load it into BigQuery or ClickHouse — the kind of scale that backs up "I've worked with big data."
A live feed of Wikipedia edits as they happen, over Server-Sent Events. The data is flowing right now — perfect for your first taste of real streaming.
If the "serious" topics pull you in, this is where clean data meets ready-made examples of turning numbers into a story.
Health, climate, economy, population — by country and year, with polished visualizations and a full data catalog to pull from. It teaches you to tell a story with numbers and to show uncertainty honestly.
Clean datasets behind their stories on sports, politics, and culture — all posted on GitHub. The trick: each one ships with the article that used it, so you can reproduce the analysis and then dig deeper.
Global economy and development by country: GDP, population, education, energy — hundreds of indicators spanning decades. It comes with a friendly API, so you can pull the data straight into your code.