I build the unglamorous half of streaming products: ingest, transcoding queues,
delivery and the metrics that tell you when any of it is quietly failing.
Fourteen years in, mostly Go and Python, mostly systems that are not allowed to stop.
Selected work
Segmenter for a regional broadcaster
Live HLS packaging with per-renditions failover. Cut rebuffering
on mobile by roughly a third; ran two years without an unplanned outage.
Ingest queue rewrite
Replaced a cron-driven pipeline with a work-stealing scheduler.
Median time-to-publish went from 40 minutes to under 4.
Storage cost audit
Lifecycle rules and a smarter rendition ladder for an archive
of ~180 TB. Same catalogue, 38% less storage spend.
Tools I reach for
Go · Python · PostgreSQL · ffmpeg · nginx · Prometheus · Linux
Writing
Why your p99 is a story about queues, not servers
Notes from chasing latency in a media pipeline.
Backpressure, explained without the diagrams
What actually happens when a consumer falls behind.