Rishta Studio
A scraping, OCR and extraction pipeline that turns image-only listings into a searchable register.
- Category
- Hobby
- Type
- Website
- Year
- 2026
- Stack
- Python · Playwright · Apple Vision OCR · Gemini · PostgreSQL · Neon · React · Vite · Express · Vercel
Rishta Studio is a research project on a Facebook page where matchmaking listings are posted as images, not text. That makes them impossible to search, sort or reason about. This turns them into structured data.
The pipeline
One entry point runs the whole chain: scrape the newest listings, OCR them, extract fields, upsert to Postgres. The grid scraper is incremental Playwright running headless, OCR goes through Apple Vision locally with a Gemini pass for the images Vision struggles with, and field extraction is regex over the OCR text. A launchd job runs it daily.
The web UI is a Vite and React app, served by Express locally and as a serverless function on Vercel, reading from Neon Postgres.
The register itself is password protected, and the interesting engineering is all upstream of it: getting reliable structure out of low-quality photographs of text.