Agentic Ticket Triage and Response

In enterprise environments, IT support teams are increasingly burdened by the volume and complexity of service desk tickets driven by expanding digital infrastructure and remote work. This project explores the development of an agentic AI framework for automating first-line IT support tasks, including ticket triage and response generation. Leveraging Large Language Models (LLMs), synthetic data generation, and frameworks like LangGraph, the system replicates core elements of IT Support Management environments. A multi-agent architecture coordinates tasks such as problem diagnosis, classification, and solution generation using retrieval-augmented generation (RAG) and semantic search techniques. Experimental evaluations benchmark the performance of fine-tuned LLMs against traditional machine learning methods like XGBoost. Results show that while fine-tuned LLMs outperform in structured classification tasks, synthetic data limitations and LLM inconsistency pose challenges in response reliability. The study highlights the trade-offs between fluency and factuality, and suggests that agentic systems, when paired with human oversight, offer a promising pathway for scalable and efficient IT support automation.