Skip to content
Haavn' Fun · Past
0825 March 2026

Working with data

Getting useful answers out of raw datasets without writing SQL. Three public datasets, Claude Code as the front-end, and natural-language questions that find the story hiding in the data.

Ask the data questions like a journalist

Most people open a CSV, get overwhelmed by 160,000 rows, and give up. This session showed that you don't need to know SQL — you need to ask the right questions, the way Michael Lewis does. Attendees picked one of three public datasets (Manhattan 311 noise complaints, every street tree in San Francisco, 80,000 UFO sightings since 1949), dropped it into Claude Code, and started asking in plain English. The AI wrote the SQL, ran it, and returned the answer. The real skill isn't the query — it's asking like a journalist: what's the story hiding in this data?

Tools covered

  • Claude Code — the primary driver of the live demos. Loads CSVs into a local SQLite database and answers natural-language questions by writing and running SQL.
  • Anti-Gravityantigravity.google. Google's answer to Cursor; a coding-focused IDE. Used in-session by attendees on PCs. Less forgiving than Claude Code but functional.
  • Cursorcursor.com. AI-native code editor. Sebastian notes Claude Code can do everything Cursor can; Cursor appeared because a prior workshop asked for it.
  • Gemini — the right tool for small datasets (under ~1,000 rows). 1M-token context window holds a CSV in memory; just upload and ask.
  • SQLite — the embedded database the AI loads each CSV into. No server, no setup.
  • Python + matplotlib — the standard library stack for producing charts from a database. Sebastian's "magic spell" — saying use Python and matplotlib shortens the path to a good visualisation.
  • Leaflet — the mapping library of choice for plotting data onto interactive maps.
← All workshops