Building a Recommendation Engine Without Slowing Delivery

This session follows how Katerina is leading the shift from rule based curation to a machine learning driven recommendation platform across the home feed and vertical video feed. She shares how the team moved out of a limbo state, shipped simple ranking models fast, and used metrics and offline LLM support to guide smarter long term platform bets.
Outcomes:

• Identify when rule based personalization has peaked and where to introduce recommendations first
• Design an incremental ML roadmap that delivers quick wins while a shared platform is still being built
• Balance heuristics, classic ML and LLM assisted training to improve feeds without bloating production
• Align engineers, product and leadership around KPIs so experiments, platform work and delivery stay in sync