Skip to content
lucaskrantz.se

Project

Hylite.dev

Real-time Twitch clip monitoring system exploring how early clip momentum can be detected.

Year
2025
Role
Full-stack engineering, data, infrastructure
See project live

Hylite started from a technical question: how can you detect early momentum before a clip becomes popular? The project monitors live Twitch streams, collects clip activity, and looks for unusual changes in clip creation over short windows of time.

Architecture

  • Next.js App Router frontend with TypeScript throughout
  • Node.js workers for polling live streams and processing clip events
  • Prisma / PostgreSQL for persistent storage and query layer
  • Redis for rate limiting and transient state
  • Analytics pages with clip snapshot tracking

Infrastructure

  • Dockerized production deployment on Hetzner VPS
  • Caddy reverse proxy with automatic HTTPS
  • Cloudflare for DNS and edge caching
  • Structured logging and basic production monitoring

What it required

The hard part was not fetching clips. It was deciding when clip activity looked meaningful without making too many requests to upstream APIs. That meant thinking carefully about rate limits, polling intervals, backoff strategies, and how workers coordinate shared state.

The project became a useful way to learn about real-time backend architecture, third-party API constraints, and the tradeoffs involved when a system depends on data that only exists behind rate-limited endpoints.