Patricija Žemaitytė is a product manager at Oxylabs building public web-data infrastructure that connects AI systems to current information. Her work focuses on the retrieval bottlenecks that determine whether models can use live search, video, and other public data reliably at production scale.
Before taking responsibility for SERP and LLM scraping products, Žemaitytė led teams working on Web Scraper API and core services. That progression from engineering-adjacent leadership to product strategy shaped her emphasis on practical constraints: changing customer requirements, browser overhead, unreliable websites, and realistic production testing.
An early customer request for large-scale video collection grew into a multimodal video pipeline spanning downloads, transcripts, subtitles, search, metadata, and channel information. When transcript requests failed because the customer actually needed subtitles, the distinction exposed how easily infrastructure breaks when technical specifications miss the format a real workflow requires.
- Search retrieval should fit the application. Conventional search scrapers collect advertisements, widgets, and complex page layouts; interactive AI systems primarily need organic results, news, and topical information. Oxylabs’ Fast Search API narrowed that scope and reworked browsers, proxies, sessions, and parsing, reaching a reported average latency of approximately 550 milliseconds. Subsecond retrieval makes live search practical inside interactive AI workflows.
- Production resilience cannot be inferred from a prototype. One low-latency search implementation worked in testing but was blocked during a customer call. Recovering required balancing the reliability benefits of browser execution against its cost, complexity, and latency.
- Realistic load testing must account for routing, rendering, proxies, retries, parsing, and delivery. Scaling Web Unblocker exposed the limits of synthetic traffic and showed that logs and metrics themselves become significant infrastructure workloads.
