~/home / lab / tagharvest

TagHarvest

Petani AI

Finds true viral outliers, never just big accounts.

Sorting short-form content by views only ever surfaces one thing: accounts that were already big. TagHarvest scores every post against its own account’s median performance and follower ratio instead, so a small creator pulling 40x their baseline outranks a celebrity having a normal day.

That is what an outlier actually is, and it turns out to be exactly what a content team wants to study.

  • Two-stage hashtag discovery finds where a niche actually lives.
  • Every post gets a leave-one-out outlier score against its own account history.
  • Results rank by how anomalous the performance is, never by raw reach.

Built in one evening sprint. First live run scanned 60 posts in 11 seconds, flagged 23 outliers with the top one at 38.9x baseline, and cost 23 cents in API calls. The write-up is in the journal.

$ ls ./evidence/ — 2 items