Jude Llm Based Representation Learning For Linkedin Job Recommendations

LinkedIn’s JUDE platform uses fine-tuned language models to turn job descriptions, profiles and resumes into numerical representations for matching. Building on the earlier Pensieve platform, it replaces standardized input features with representations learned from text.

The expensive model runs when relevant information changes; ranking systems then read stored results when serving recommendations. LinkedIn says updates are reflected within seconds and reports a 2.07% increase in qualified applications after deploying JUDE embeddings in job recommendation and search ranking. That is an application metric, rather than a measured increase in completed hires.