How We Built Domain Adapted Foundation Genai Models To Power Our Platform

LinkedIn developed EON models to reuse professional-domain training across several AI features. The models learn from examples covering multiple tasks and their reasoning, followed by preference and safety alignment. One application is Hiring Assistant's candidate evaluator, which reads profiles, résumés, recruiter notes and job requirements, then returns a match assessment with an explanation. LinkedIn reported that this evaluator accounted for nearly 90% of language-model calls in that workflow, and said the model behind it must be fast and accurate at that scale. In LinkedIn's own testing, the EON-8B model supporting it matched candidates to job requirements more accurately than GPT-4o mini by an absolute 4%; the post does not publish the evaluation details.