Natural Language Driven Leave Management Using Lang Chain Agents and Few-Shot Learning

The traditional approaches for managing leaves involve form-driven procedures and processes, resulting in high levels of admin workload, delays, and low levels of flexibility for users. In this paper, we introduce a novel Natural Language Driven Leave Management System where employees are able to ask their requests about leaves through natural language. The introduced framework uses the capabilities of LangChain agents, Gemini large language models, few-shot learning, and MongoDB to translate natural language into NoSQL operations. A mechanism of validation and fallback is used to guarantee schema, data consistency, and query reliability. The framework architecture includes four layers that enable to implement automation of leave processing and notification management. An experiment was performed for testing the framework by applying a number of representative queries in leave management. Configurations include Zero-Shot, Few-Shot, and Few-Shot with Validation cases. According to the results obtained, the developed framework achieved Intent Accuracy of 91%, Field Extraction Accuracy of 88%, NoSQL Query Accuracy of 87% and Fallback Success Rate of 93%.