Movie Rails: Tell Me What You Think

A few weeks ago I wrote up how my capstone team built Movie Rails — clustering 8,000 films into 67 algorithmically-discovered streaming categories with UMAP, HDBSCAN, and SHAP. That post explained how the dragon got slain.

This post is shorter. Movie Rails is now live.

The full Streamlit dashboard — every rail, every film, every feature badge, every poster — is hosted on Hugging Face Spaces. You can browse the rails, click into any cluster, see the SHAP-derived feature profile, and rename or merge categories the way a human curator would.

Open the live demo (it runs on a free Hugging Face CPU instance — first load can take ~30 seconds while the container warms up).

What I'm hoping you'll do

• Click around. Browse the 67 rails. Find one that feels off. Find one that surprises you.
• Read the feature badges. Each rail's top SHAP features are listed — that's what makes the cluster distinct. Sometimes they're obvious ("heist," "ensemble cast"). Sometimes they're weird in a way that makes you go oh, that's actually right.
• Try the curator tools. You can rename rails, disable bad ones, and move films between categories. This is the human-in-the-loop layer that I think is the project's most important design choice.
• Tell me what's broken. Honestly. The clustering uses content features only — no audience signal — so some rails are probably nonsense. I want to know which ones.

How to send feedback

Easiest path: drop a comment below. Or, if you'd rather skip the form, use the contact form on my site or hit me on LinkedIn. If you found something specific — a weirdly-named rail, a film that's clearly in the wrong cluster, a UI bug — screenshots help.

If you want to dig into the technical side, the capstone write-up covers the full pipeline. The Hugging Face Space is the same code, just packaged as a public, runnable demo instead of a Jupyter notebook.

---

Built with Jordan Ehrman and Manuel Lara as part of the MSBA Capstone at Chapman University. Hosted on Hugging Face Spaces because $0/month is a great price.