analyze your AI usage by seeing what you use AI for the most and how you use AI over time
currently have:
- cluster conversations to see how you have used at a point in time
- see you many messages you send over time (day, week, month, year)
will have:
- see what kind of messages you send over time
- hierarchical clusters
- update online
some concerns:
- HDBSCAN classifies too many messsages as noise
- KMeans doesn't classify well/LLM doesn't name clusters well
- LLM naming won't work when clusters are super big
ideas:
- instead of machine finding clusters, first cluster according to intent categories then classify by topic
- intent categories: asking, doing, expressing
useful resources: