AI-DRIVEN AGENTIC FRAMEWORK FOR ENTERPRISE SYSTEM TROUBLESHOOTING WITH ENHANCED RAG MODELS

Abstract

Technical troubleshooting in enterprise settings frequently requires navigating various, heterogeneous data sources to effectively resolve intricate issues. This paper introduces an innovative agentic AI solution based on a Weighted Retrieval-Augmented Generation (RAG) Framework designed for enterprise technical troubleshooting. By dynamically weighting retrieval sources, including product manuals, internal knowledge bases, FAQs, and troubleshooting guides according to query context, the framework prioritizes the most pertinent information. It prioritizes product manuals for SKU-specific inquiries while integrating general FAQs for more extensive concerns. The system utilizes FAISS for efficient dense vector search, along with a dynamic aggregation mechanism to integrate results from various sources seamlessly. A LLaMA-based self-evaluator guarantees the contextual precision and assurance of the produced responses prior to their dissemination. This iterative process of retrieval and validation improves precision, diversity, and reliability in response generation.Initial assessments of extensive enterprise datasets indicate the framework's effectiveness in enhancing troubleshooting precision, decreasing resolution durations, and accommodating diverse technical challenges. Future research intends to improve the framework by incorporating sophisticated conversational AI functionalities, facilitating more interactive and intuitive troubleshooting experiences. Efforts will concentrate on enhancing the dynamic weighting mechanism via reinforcement learning to optimize the relevance and accuracy of retrieved information. By integrating these advancements, the proposed framework is set to transform into a holistic, autonomous AI solution, revolutionizing technical service workflows within enterprise environments.