Engineering LinkedIn's Hiring Assistant for International Markets

This post details the engineering effort to internationalize LinkedIn's Hiring Assistant for French and German markets. It addresses the challenges of on-the-fly text generation in agentic AI systems, which differ from traditional software localization. Key technical contributions include developing a shared rubric framework to capture language-specific rules (gender agreement, formal/informal address, noun capitalization, number/date formats) and training per-language Low-rank Adaptation (LoRA) adapters for smaller, cost-efficient models. This approach allows for native quality generation without multiplying serving costs, enabling the Hiring Assistant to feel native in new markets.