gpu-retrieval

This paper introduces a policy-aligned retrieval framework for semantic search on LinkedIn, leveraging embeddings partitioned into category-supervised segments and a two-stage GPU architecture to improve recall and precision, with significant gains validated in A/B testing.