There are some practices that add value. There are others which rather extract value. This project attempts to walk a fine line between the two.
Early focus is on automated value extraction from Twitch clips though. We download popular Twitch content, classify it and group it. Then we export video files ready to be edited together into a final video which can be uploaded to YouTube.
apt install protobuf-compiler libprotobuf-devSqlite connection behind a mutex.
The convention is to use reference to the connection for scopes which don't yield (or not for very long anyway). Scopes which yield take the mutex and lock it as they need. This allows reasonable concurrency in exchange for simple management.
We typically run first RPC and then insert to the database. It's more likely that RPC fails than the database op.
Simple no-css handlebar templates.
- label data with whisper and gpt
- given context and clip, cluster clip with respect to context
- the transcribed text is then vectorized
- search on vectors and select the top N clips
- given context and clip, return a fragment that can be used in the final edit