Moment Search: A Searchable Video Archive Built with Claude Code
Since September 2020, the Techspressionism community has met online almost every month. Artists from dozens of countries log on, show work, and argue about what technology does to art and to us. Those conversations were recorded and posted to YouTube. By this fall there were 149 recordings, about 220 hours of video.
That is a lot of good thinking that nobody could find. If you remembered that a curator said something sharp about AI in a salon in 2022, your only option was to scrub through a two-hour video and hope.
So I came up with a search engine for it, and Claude Code built it.
What it does
The Techspressionism Video Archive lets you type a word and see every place it was spoken across all 149 recordings: 104 salons, 30 artist interviews, 6 roundtables and 9 presentations, from 2020 to 2026.
Each result is a moment search result: a link that opens the video at the exact moment the words were said. Not the start of the video. The second.
Here is one. In the first Techspressionism roundtable, Curators in Conversation (October 3, 2022), the Whitney’s Christiane Paul and art historian Helen A. Harrison talk about the difference between technology as a tool and technology as part of the artwork. This link starts right there, at 26:39.
That’s what I mean by moment search: searching a video collection like a book, where every answer is a link to the moment it happened.
How it works, in plain terms
- Transcripts. When a Zoom transcript exists, the archive uses it, because Zoom labels who is speaking. Otherwise the audio is transcribed with Whisper, an open speech-recognition program, and YouTube captions fill any gaps. The timestamps are matched to the YouTube videos, even ones that were trimmed before upload.
- Honest names. A quote is credited to a person only when Zoom’s label, the video description or a listener’s confirmation backs it up. Anything else is marked “Unattributed” and left out of the results.
- Citations. Every result has a Copy Citation button: speaker, recording, date, timestamp and link.
- Artist pages. Each artist has a page showing where they speak and where others mention them.
- Open data. There’s a downloadable list of recordings, a text transcript for each one, and a file written for AI assistants (llms.txt) so they can find and quote the archive accurately. The transcripts are free to reuse with credit (CC BY 4.0).
- 13 languages. The site’s menus and pages work in Spanish, French, German, Chinese, Japanese, Arabic and more.
- A permanent, citable copy. The whole archive is deposited on Zenodo, the open research repository run by CERN, with its own DOI: 10.5281/zenodo.23990824. A DOI is the permanent ID scholars use to cite papers and datasets, so researchers, curators and students can cite the archive itself, and that link will keep working even if a website moves. The deposit is free to reuse with credit (CC BY 4.0), and the code is on GitHub.
Machine transcripts make mistakes, especially with names and art terms, so the archive says so plainly and asks you to check a quote against the video before citing it.
Built in Python, with Claude Code
I’m a painter, not a software engineer. I conceptualized the archive, and Claude Code, an AI coding assistant, coded it in Python. I described what I wanted, it wrote and ran the code, and I tested, pushed back and made the calls. That loop let one artist make something that would have needed a small development team a few years ago.
It also fits what Techspressionism has been about since I first used the word in 2011: using technology to express something human. Here the technology is a search box, and the human part is 220 hours of artists explaining why they make what they make.
Are there others like it?
A few. PBS’s American Masters Digital Archive puts full, unreleased interviews online with synchronized transcripts you can search and click to jump to a moment. C-SPAN’s video library lets you search everything said in Congress and clip it. Tools like Filmot search the captions of YouTube videos, and Google shows “key moments” inside some videos in search results.
What I haven’t found is moment search built by and for an art movement: a community’s own memory, searchable to the second, with every voice credited and every quote citable. That’s the part I care about.
Try it
Search a word you care about, like AI, painting, color or memory, in the archive. Then send someone the moment.
Related: The Art Meme, my 2005 essay on how the idea of “art” spreads from mind to mind.