Analyses of the Dunc'd On Basketball Podcast

A demo of the production app predicting the best episodes

production_duncd_on_episode_model.mov

The Dunc'd On Podcast only has one problem: they put out too much good content.

So, I build an app that would predict which recent episodes I'm most likely to enjoy.

This production app hits the Dunc'd On RSS feed for episodes over the past week, performs feature engineering, and runs a model that achieved an F1 score of 0.80 for predicting the best episodes (as labeled by me). You can try the app yourself here.

A demo of exploring episode description embeddings

Screen.Recording.2025-01-14.at.4.50.37.PM.mov

Progress so far

Hello! So far I've created a couple of data visualizations showing:

• Nate and Danny release episodes at the wildest times of day
• A way to determine how similar or different episodes are based on embeddings from ModernBERT. Now with episode types included and an interactive element (see video example above)
• A breakdown of whether I thought the episode was a banger by the type of episode (Those features were created using the data labeling app I talk about below)

I've also created a labeling app using Shiny for Python where I can categorize each episode and designate whether or not I think the episode was a banger. I'm labeling episodes using this tool in service of creating models that can categorize the kind of episode and predict whether I'll think it's a banger or not.

The current culmination of this repo is the production app described in the first README section!

I may do more development on this project in the future, so if there's something you'd be interested in feel free to file an issue or post at me on Bluesky