Human resource (HR) departments in small and medium enterprises face challenges such as high operational costs, regulatory compliance, and routine task management, compounded by limited computing resources and data privacy concerns. To address these issues, we introduce a lightweight, on-premises language solution using a fine-tuned TinyLlama model integrated with a retrieval-augmented generation model for HR applications. Leveraging parameter-efficient methods, such as low-rank adaptation, the model shows excellent performance with a single graphics processing unit. The retrieval system is accurate in accessing local legal documents, complying with Malaysia’s regulations, while preserving data sovereignty. This approach provides SMEs with cost-effective, transparent, and scalable HR support.
Wong et al. (Fri,) studied this question.