Peter Clark

Product Manager and tinkerer.
Writing about product management4, productivity3, cars2, 3d printing1, 3d prints1, brio1, diy1, lego1, living1, music1, projects1, reviews1, setup1, steve jobs1, toys1 and writing1 in the format of text posts13, embed posts3, link posts2 and thought posts1. There are even draft2 posts.

Lego Set Findr

It's a little difficult to find lego sets because of their annual release schedule. The sets you see on the Lego site (or store) are the most recent ones, but often there are some real gems that were released just last year that are no longer easily surfaced.

I built a simple (ish) tool that allows you to find complementary lego sets to lego sets you enjoy.

This was a really fun project, and not as easy as you might think — if you deploy a simple collaborative filtering model it simply finds models that have similar names: so police station sets return other police station sets.

What I wanted was complementary sets: police stations should return fire station sets.

I achieved this with rather elaborate logic:

  1. I weighted sets released within 3 years of the parent set more highly, and sets released the same year slightly lower.

  2. I weighted sets with the same theme much more highly.

  3. I weighted sets with a similar number of pieces slightly more highly.

  4. I asked ChatGPT to provide 1-10 complementary lego sets for each set, these are ranked highly.

  5. I weighted sets with a very similar name significantly less highly. For example "Police Station" won't rank "Police Station 2.0" very highly at all.

One of my test samples is Lego Grocery Store — the highest ranked complementary set is Lego Apartment Building which would not score highly if not for this elaborate logic.

I imported lego sets with over 20 pieces and that were released after 2000, so the data set is fairly comprehensive. Feedback is welcome, although I recognize this is all a bit esoteric.

Check it out!