The book
Do you really need a cloud warehouse?
Most teams query gigabytes, not petabytes — and pay warehouse prices for laptop-sized work. Local-First Analytics shows how to run warehouse-class analytics on hardware you already own with DuckDB, Parquet and Arrow: faster feedback, zero lock-in, a fraction of the bill.
314 pages · 12 chapters · runnable code for every chapter · revised August 2026
The Warehouse Reality Check
Six questions, scored in your browser. No email, no login — your answers never leave this page. It tells you whether your workloads belong on a warehouse, on your laptop, or split between the two.
The same diagnosis I run in paid audits, reduced to its bones.
One stack. Eight layers. Add them as you need them.
The book builds a complete analytics platform in the order you'd actually adopt it — files first, everything else optional until the data demands it.
- 1Parquet — one open file format as the source of truth
- 2DuckDB — warehouse-class SQL on your laptop
- 3Arrow — zero-copy data movement between tools
- 4Polars — transforms that embarrass pandas
- 5Pandera — quality gates that fail loudly
- 6Prefect / Dagster — orchestration when cron stops scaling
- 7Evidence / Metabase — BI without the bloat
- 8Ollama — ask your data questions — fully offline
Why I wrote it
I've spent close to two decades building data platforms — including the engineering backbone behind stock analysis at Stockopedia — and I kept watching the same scene: a team renting a cloud warehouse, at four or five figures a month, to query data that fits on a laptop.
So I wrote the book I kept wishing I could hand them. It's the playbook I use in my own consulting practice: keep truth in open files, query them at warehouse speed, and add infrastructure only when the data — not the vendor — demands it. Every claim in it ships with code you can run.
Get the book
Kindle and paperback, on Amazon. The price is less than ten minutes of your warehouse bill.
Or request a free review copy
I set aside review copies for people who'll genuinely use the book — students, open-source maintainers, and teams weighing a move off their warehouse.
Tell me a little about your stack and I'll read it myself — no auto-responder. If it looks like a fit, the PDF lands in your inbox within a day or two.
Questions people ask
Do I need to be a data engineer to read it?
No. If you can write basic SQL or Python you can follow every chapter. The book was written for the person who has data and questions but no platform team — each chapter builds one layer, and every layer is optional until you need it.
Will this actually replace Snowflake or BigQuery?
For a lot of teams querying gigabytes — yes, entirely. For teams at TB scale with real concurrency, it replaces the expensive parts: development loops, exploration, quality checks and most BI. The book is honest about which is which; take the reality check above for your answer.
What exactly is in the book?
314 pages across 12 chapters and an appendix: the local-first stack (DuckDB, Parquet, Arrow, Polars), performance mechanics, partitioning, data quality with Pandera, orchestration from cron to Prefect and Dagster, versioning data like code, BI with Evidence and Metabase, and local AI with Ollama. Revised August 2026 for current library versions.
Is the code available?
Yes — every chapter ships with runnable scripts and notebooks in a public GitHub repository, organised one directory per chapter, with a Docker Compose dev stack included. You can run everything in the book without buying anything.
Can I still get a free copy?
You can request one below. I read every request personally and email the PDF to requests that look like a genuine fit — usually within a day or two. It’s a review copy, not a download link: I’d rather send it to a hundred people who’ll use it than have it sit in ten thousand download folders.
Your laptop is faster than you think.
Get the book on Amazon