Tango Video Search

Problem: YouTube search for tango performances is inefficient. Results are buried due to YouTube's algorithmic sorting and inconsistent metadata.

Solution: Tango Video Search provides a streamlined interface to filter by dancer or orchestra with minimal friction.

How it works: It’s powered by an LLM-curated index built from YouTube metadata. That means faster, more accurate retrieval of the performances you're looking for.

Features

• Filter videos by dancer combinations
• Filter videos by orchestras
• Responsive video grid layout
• Quick filtering through video cards
• Video metadata display

Tech Stack

• Framework: React Router v7 (started as a Remix project)
• UI: Radix UI + Tailwind CSS
• Database: SQLite with Drizzle ORM
• Deployment: Fly.io

Development

To run the development server:

npm run dev

Database

Note

The database is not included in the repository.

The application uses SQLite via Drizzle ORM with a database file at data/sqlite.db by default, or override the path using the DATABASE_URL environment variable. The schema includes:

• Videos (YouTube metadata)
• Performances (tango-specific metadata)
• Dancers, Orchestras, Songs, Singers, and Curations

Deployment

Deploying the application is handled by GitHub Actions, which will automatically deploy to Fly.io when you push to the main branch.

Updating Data in Production

After preparing a new database file:

1. Copy the .db file into /data and update the sqlite.db symlink:

   ln -s sqlite-2025-MM-DD.db sqlite.db

2. Connect to the Fly.io instance and remove the old database:

   fly ssh console
   rm /data/sqlite.db*

3. Upload the new database file:

   fly sftp shell
   put data/sqlite.db /data/sqlite.db

4. Restart the application:

   fly apps restart