Every 3 seconds, Rafiki 3.0 is trained on the latest threads posted on INLEO. Here's how we now have a full LLM for Threads:

1. Every 3 seconds, the HAF scanner polls the hive blockchain for new threads
2. Threads get their metadata (author, permlink, depth, timestamp) and of course, their body text indexed in a Rafiki-Trained Vector space using our hardware + local model
3. When a user searches, their query is embedded into that vector space and Rafiki semantically finds similar threads using the "nearest-neighbor" search approach
4. Results are re-ranked according to this approach with top-level threads gaining a boost, comments and low-substance posts getting penalized in the ranking and then the data is returned to Rafiki so he can summarize it intelligentally

Currently, we have 3.84M Threads indexed with new ones being added within 3-9 seconds of when they're published onchain