Marsad
Short-video intelligence for Instagram & TikTok
A platform that follows public Instagram and TikTok accounts, pulls their clips and turns each one into an Arabic analysis: summary, hook, timestamped verbatim script, on-screen text and call to action.
- Problem
- Competitors' reels were being watched by hand, and the conclusions lived in someone's memory. Understanding what makes a clip work had to become searchable data instead.
- Built
- FastAPI backend with a Go fetcher service and a TikTok scraper CLI behind it, fetching public pages over residential proxies
- Gemini for what the video shows, Whisper for what it says, and OpenAI for the Arabic narrative and structured hook, script and CTA extraction
- A job queue that runs analyses one at a time and streams progress to every teammate's browser, reattaching after a reload
- Bilingual Next.js interface with URL-driven filters, shared team state and a grounded question-and-answer thread per clip
- Impact
- Analyses are computed once and shared, so the pipeline watches the clips instead of the team.
- Role
- Architecture, backend, frontend
- FastAPI
- Go
- Next.js
- TypeScript
- Gemini
- Whisper
- OpenAI
- SQLite
Private deployment · no public link