Social scraper & reply
An agent system that researches, tracks, and drafts content for @deepika.builds across Instagram, YouTube, Reddit and X, built before Claude could drive a browser, so it scrapes instead. It has since grown into the full content engine. The flagship case study lives at /work/content-engine.
A social media research and analytics agent. It scrapes trending AI/dev content and pulls apart the format and hook patterns, tracks specific creators for what's working, monitors AI discourse on Twitter and Reddit for angles, analyzes who's actually engaging with my reels, and pulls performance data to find patterns in what lands.
Every week it produces a batch of seven reel briefs: the format, exactly what to film, the text overlay word-for-word, and a ready-to-paste caption. Each brief has to be producible in under ten minutes or it gets killed.
Four Apify scrapers wired in over MCP: Instagram, YouTube, Reddit, and X. Everything flows through Notion: an ideas dump as the single source of truth, and a shared activity log that my other agents write to as well, so content ideas come from work that actually happened.
The important rule: nothing gets scripted until I've filled in my note on the idea, at least a couple of lines with my angle. The agent researches and drafts; the opinion has to be mine.
The system encodes the content strategy as hard constraints. Every piece must pass the DM test: would someone send this to a friend? Every hook makes exactly one promise and the body delivers that promise with one example; if it needs multiple points, it becomes a carousel instead. Reels cap at 40 seconds and 30 minutes of production time. Content must come from what I already did, never build something just to film it.
The visible output is the carousel system: typography-forward slides, serif headlines, one accent color, generous whitespace, generated as HTML/CSS and rendered to PNGs. A few covers from recent batches below.
