I wrote a little Chrome extension to make animated WebP (“weppy”) files from a region of a webpage:

https://chromewebstore.google.com/detail/weppysnap/kepcjkdedeocincdhenhepinpkkoghmn

I use it when writing documentation and I want to show a short animation of how to do something (great for Github README for example). A bit simpler than a WebM video and more modern than an animated GIF.

I'll try to port it to Firefox at some point.

Example and analysis of how AI web scrapers are breaking small and medium cultural heritage sites.

AI Scrapers vs Wikibase
https://semlab.io/blog/ai-scrapers-vs-wikibase.html

- Analysis of 13 million requests to our Wikibase over 2 weeks
- Good vs Bad bots (self-identifying vs camouflaged)
- Close look at how GPTBot interacts with the Wikibase instance

Blog: Visualizing 14,000 Released Epstein Emails.

I built a viz of the emails released as part of the 20K House Oversight Committee docs.

https://thisismattmiller.com/post/email-visualization/

- Provides a clustered high level view of the emails exchanged by contact across time
- Allows you zoom into individual email exchanges and open the source documents

Future shape of the Library of Congress?

https://www.congress.gov/bill/119th-congress/house-bill/6028/text

Librarian appointed by congress, Copyright office removed from the library.

Blog Post: LCNAF & Trie – Storing +11M unique names in 50MB data structure in the browser

https://thisismattmiller.com/post/lcnaf-trie/

- Optimizing LCNAF authorized headings into a trie data structure
- In browser MARC file (binary & XML) name reconciliation tool
- In browser LCNAF search tool
OpenRefine / API / Command line tools for name reconciliation

Looks at what applications are possible when you can represent large indexes like LCNAF in a small data footprint.

Halloween blog post: Italian Giallo Horror Films

https://thisismattmiller.com/post/giallo/

- Using vision language model Qwen2.5-VL to analyze a 70 film corpus (🧟) / 80,000 frames
- Automatically build “trope clusters” finding similar scenes across movies
- Plotting tropes across the run time of movies to see patterns

If nothing else probably the longest eye acting supercut you’ve ever seen: https://youtu.be/cGrmkOwut6k

New Post: PEN America Banned Books 2025 dataset

https://thisismattmiller.com/post/book-bans-2025/

Looking at school district book bans

- Interactive Map interface to the books banned in 2024-2025
- A faceted browse interface to the 3700 books
- Subject heading analysis

New Blog Post.
Library of Congress & Flickr Commons: Analysis of user interactions on 40,000 images
https://thisismattmiller.com/post/lc-flickr-commons/

- Organizing 95K photo comments by embedding clustering.
- Viewer to explore user georectified images
- Folksonomy tagging vs LCSH Vocabulary
- Placing into the Wiki* knowledge graph

Trying out workflows that use multimodal LLMs for validating and QA.

In this blog I walk through a test using 1000+ Siskel and Ebert videos to extract key video frames and other data.

https://thisismattmiller.com/post/building-datasets-from-video-collections-using-local-cloud-llms/