Probably not news for api touchers, but I found a repo by @why.bsky.team that implements a bunch of simple and less simple feed generators, including ones using ML to classify posts.
Pretty straightforward. Wonder how the ML models keep up with firehose throughput though
https://github.com/whyrusleeping/algoz
Pretty straightforward. Wonder how the ML models keep up with firehose throughput though
https://github.com/whyrusleeping/algoz
Comments
Then again, what about our savior Kotlin Multi Platform? ;)
https://github.com/whyrusleeping/algoz/blob/master/classify/main.py
At 100 posts/sec from the firehose, I wonder how many contain images. Maybe it's not that much data after all, and a single ViT instance can handle it.
we only see around 15 images a second most of the time, so it keeps up pretty well
TODO: Look at this Bluesky feed generator repo for good examples