Recall Models
Built a high-throughput retrieval system handling 1K+ QPS for e-commerce search, retrieving relevant products, videos, and LIVE shopping streams for personalized queries and feeding candidates into downstream ranking models.
Relevance Models
Developed SLM-based relevance models for e-commerce search ranking to predict TikTok Shop product relevance through deep semantic understanding of query-item similarity, surfacing products that better match user search intent.