Two things I am building for myself.

Every night it collects recent public posts from around 150 tracked accounts through a data provider, computes an engagement-over-velocity score, and promotes the outliers into an analysis queue. What comes out the other end is a structured brief: the mechanic that made it work, the hook family it belongs to, and an adaptation for each brand voice.

It is a FastAPI service with a Postgres store, deployed with scheduled jobs so it keeps running whether or not anyone is at a desk. The interesting part is not the collecting, it is the scoring: raw view counts reward accounts that were already big, so the ranking has to normalise for reach and reward velocity instead.

The problem with running a distributed content operation is that the truth lives in six places at once. Production status in one tool, publishing state on the devices, revenue in a platform export, spend on a bank statement.

This project pulls those into a single read-only surface with one job: make the gap between planned and actually shipped impossible to miss, and put revenue next to the spend that produced it.