TechPrufer
Sign inGet started

Blogs›LinkedIn›Semantic Search Stack

Semantic Search Stack

Semantic Search Stack

2
posts
2026

LinkedIn has developed a new semantic search stack leveraging Large Language Models (LLMs) to provide a more intuitive, personalized, and predictive search experience. This involves transforming the search infrastructure to incorporate LLM-based components for query understanding, embedding-based retrieval (EBR), and cross-encoder ranking. The system aims to move beyond keyword matching to interpret natural language and infer user intent, improving relevance and efficiency at scale. Key technical advancements include the development of the MUSE (Member Understanding Semantic Embeddings) platform, which uses a dual-tower Siamese architecture with Matryoshka embeddings for efficient retrieval and ranking. The MUSE Teacher, an open-weight reasoning model, was developed to generate large-scale relevance labels by replicating an expert LLM judge, improving the quality of supervision for the embedding model. This system is applied to LinkedIn's Hiring Assistant to semantically match recruiter queries with candidate profiles.

2026

Semantic Search for AI Agents at Scale: Retrieval and Ranking for LinkedIn’s Hiring Assistant

6/11/2026

This post details the engineering effort behind the semantic search strategy for LinkedIn's Hiring Assistant, focusing on the MUSE (Member Understanding Semantic Embeddings) platform. It describes the MUSE Teacher, an LLM-as-a-judge system for generating large-scale relevance labels, and the MUSE embeddings architecture, a dual-tower Siamese model using Matryoshka embeddings for efficient retrieval and ranking. The post highlights the challenges of mapping natural language queries to candidate profiles at scale and the technical solutions implemented, including query understanding, embedding-based retrieval (EBR) using approximate nearest neighbor (ANN) search, and blending with other retrieval strategies for a final re-ranking by an L2 ranker. It also discusses the importance of a shared constitution for defining 'qualified' and the iterative prompt engineering of the MUSE Teacher.

Reimagining LinkedIn’s search tech stack

1/21/2026

This post details the engineering effort behind reimagining LinkedIn's search tech stack with LLMs. It introduces a semantic search infrastructure that uses query understanding modules, embedding-based retrieval (EBR) on GPUs, and a Cross-Encoder Small Language Model (SLM) for ranking. Technical innovations include optimizing LLM inference with score caching, ranking-depth control, and traffic shaping. A hybrid inference pipeline using Spark, Flyte, and Flink generates features and job representations. The post also describes a novel approach to product policy relevance measurement using LLM judges, including prompt engineering, distillation into smaller models, and continuous quality monitoring workflows. The EBR model is trained using a dual-tower architecture with a contrastive InfoNCE loss and a margin-based ranking loss, incorporating hard positives and negatives.