Meta Superintelligence Lab:

• LLM post-training for deep search: designed and built a synthetic data generation pipeline for complex multi-hop search questions; implemented context management for long-horizon agent workflows; and designed reward functions for agentic RL training, improving BrowseComp performance by X%.

• Led end-to-end retrieval and ranking model development for news search, launching across all English-language Meta AI users and improving DAU by X%.

Meta Superintelligence Lab:

• Build async caching system for query understanding, scaled LLM based QU signals to all Meta AI queries.

• Build freshness dedicated funnel for fresh search experience.

Modern Recommendation System