Our feed algorithms are pretty simple

1. Retrieve the posts for all members in a list
2. Include posts with matching keywords

We can easily improve this by programmatically discovering who to include. A good place to start would be
by writing an algorithm that

1. Takes a set of seed accounts
2. Walks the graph one degree (i.e. people they follow or people that follow them)
3. Apply some criterion to those list of people to decide whether to add them to the list
   • e.g. Fetch some of their posts and see if they include any of the keywords defined for that community

I'm not sure whether it makes more sense to start with looking at people the seeds follow or people that follow the seeds.
Assuming they are both equally easy to retrieve then I think starting with looking at followers of a given account might be better.

The reason is I think within communities there are "hubs" that everyone follows. For example, in devops/platform engineering I think kelseyhightower.com is one of the first accounts everyone follows. He has ~20K followers. So my conjecture those 20K followers provides high coverage of the devops community on bluesky.