Hierarchical Long-Term Semantic Memory for LinkedIn's Hiring Agent

LinkedIn has introduced the Hierarchical Long-Term Semantic Memory (HLTM) framework, a novel system designed to give LLM agents robust long-term memory for personalized, context-aware interactions. Published on arXiv (cs.IR/2604.26197), the paper details how HLTM extracts implicit and explicit signals from noisy longitudinal behavioral data, stores them in a structured schema-aligned memory tree, and supports low-latency retrieval. This addresses five critical challenges for industrial-grade memory: scalability, low-latency retrieval, privacy constraints, cross-domain generalizability, and observability.

Deployed in LinkedIn's Hiring Assistant, HLTM significantly improves answer correctness and retrieval F1 by more than 10%, while advancing the Pareto frontier between query and indexing latency. The framework's adaptation mechanism allows it to generalize across diverse use cases, powering core personalization features in production hiring workflows. This marks a major step toward more intelligent, memory-driven AI agents in real-world products.