select links from 2025-11-01

Putting ducks in orbit!

Nov 01, 2025

Hello friends! This time a short but sweet blogpost – normally I find some spooky Halloween themed blogpost but no luck this year. I mean, I could share that you can run DOOM on DuckDB but that’s from April. If you yourself wrote something, let me know - I would love to share it!

The image depicts two anthropomorphic cats in a business meeting setting, sitting at a table with papers and a maze containing two mice. One cat, wearing glasses, appears studious, while the other looks eager. The post text from owen cyclops (@owenbroadcast) references years 1972 and 1963, suggesting a humorous, timeless meme format. The caption "Fantastic data. When do we eat them?" adds a satirical twist, implying the mice are data to be consumed, blending business and predatory humor. No platform watermarks like Instagram or TikTok are present, and the line art is clear and well-executed. — owen cyclops on X

Code like a surgeon

Geoffrey Litt wrote this short blog post describing AI agents as staff members to whom you can delegate tasks. It’s a good read and the title directly references the book “The Mythical Man-Month”. I myself haven’t finished reading it but Mythical Man-Month is interesting because some parts are outdated, some parts resonate with me A LOT, other parts feel like they’re important but I feel like the ideas fly over my head. Give them both a read!

Putting ML in your database

Dirk Eddelbuettel recently released duckdb-mlpack – a duckdb extension that lets you run machine learning models on your duckdb database. Just look at this:

CREATE TEMP TABLE A AS SELECT * FROM mlpack_adaboost(”X”, “Y”, “Z”, “M”);

The idea is that mlpack, being a C++ library, integrates well with duckdb and lets you run models like Adaboost or linear regression without leaving your cosy duckdb database. While still an MVP, I love this idea - a lot of the times you don’t need a full-blown framework in R or Python to run your models. I will be keeping an eye on this extension for sure.

Another cool package is the orbital package from Emil Hvitfeldt. It works a little differently because it’s not coupled to duckdb and can work with virtually any database. The difference is that you run a machine learning model like you would do any other model but then save the model as a special orbital object. Then, when you want to to predictions, orbital sends weights/coefficients to the database for the database to compute the prediction. So instead of linear algebra, you’re just running a series of SQL commands on the database. I like this package because you could in theory save the model weights as a dbt macro because we haven’t used it yet.

High SQL

Discussion about this post

Ready for more?