How Web Data Infrastructure Powers the Next Generation of AI — Patricija Žemaitytė, Oxylabs
Patricija Žemaitytė from Oxylabs shares experiences building infrastructure for AI services at scale.
She describes three main challenges: a search API that was supposed to respond in under a second got blocked live with a client, requiring a rebuild by removing slow browsers—resulting in 550 milliseconds instead of 4 seconds. A video API request with a two-week deadline grew from a simple job into an entire product suite (transcripts, subtitling, search, metadata) when the client discovered new needs, and scaled to 30 petabytes of data without yet paying. Scale itself became a problem: when she tried to test an "unblocker" service up to 60,000 requests per second, the tests stalled at 20,000, partly because the telemetry itself became so large that it affected what she was trying to measure.
Vibekollen prepared this summary with AI from the original publication. The content belongs to AI Engineer.
