How I built an ingestion pipeline for an AI property assistant

Plan with the strongest model, implement with smaller ones

How I compress agent skills with progressive disclosure

Proof before fixing AI review findings