Case studysponsoredyt

Every sponsor read, searchable.

sponsoredyt is a pipeline and a website that capture the sponsor segments of tracked YouTube channels and make them searchable by sponsor, channel and video.

Status
Live at sponsoredyt.com.
Role
The pipeline, the public site and the admin tools.
Built with
  • Next.js
  • TypeScript
  • PostgreSQL
  • Redis
  • BullMQ
  • Ollama
  • SponsorBlock

The problem

Sponsor reads are buried inside videos and descriptions, in a different format every time. sponsoredyt turns each one into a record: brand, product, category, offer code and, for spoken reads, the seconds where the read starts and ends.

What was built

  • A public site to browse and search videos, sponsors and channels, with analytics charts and an RSS feed for each sponsor.
  • A form to request a new channel, protected by Cloudflare Turnstile and rate limiting.
  • Admin tools to add channels, reprocess videos and merge duplicate sponsors.

How it works

  • New uploads arrive by WebSub push from each channel's YouTube feed, checked against a per-channel signature.
  • A BullMQ worker on Redis gets the metadata and the transcript from transcriptyt, and the community's SponsorBlock segments.
  • For each sponsor segment, the transcript lines inside it, plus or minus 5 seconds, go to a local language model through Ollama, which returns a structured sponsor with a confidence score. A second pass reads the description.
  • Raw responses are stored, so the extraction can run again without fetching anything.
  • PostgreSQL with plain SQL migrations and full-text search; Next.js for the site.