Scientific paper proposal: XPERT: Empowering Incident Management with…

[YasmineSnowBud]

Assignment Proposal

Title

XPERT: Empowering Incident Management with Query Recommendations via Large Language Models

Names and KTH ID

• Love Lindgren (lolindgr@kth.se)
• Yasmine Schüllerqvist (yasmines@kth.se)

Deadline

• Week 4

Category

• Scientific paper

Description

We would like to present the scientific paper "XPERT: Empowering Incident Management with Query Recommendations via Large Language Models." Xpert is an AIOps system that uses large language models to recommend queries for cloud incident management.

We plan to examine how XPERT addresses the difficulty of writing domain-specific language (DSL) queries during cloud incident management. Throughout the presentation, we will discuss:

• The paper’s empirical study of Kusto Query Language (KQL) query usage
• How XPERT uses historical incident data and LLMs to generate queries for new incidents
• How the proposed XCORE metric measures query quality from the three perspectives explored in the paper
• The reported results from both the offline evaluation and deployment in a real production environment
• Highlighting the strengths and limitations of the study, possibly comparing it with two alternative approaches to AI-assisted incident management.

Relevance
Incident response is a core part of keeping software reliable after deployment, making it an important DevOps feedback loop. XPERT demonstrates how AI can be incorporated into an existing DevOps workflow and enhance the experience for its engineers by reducing the manual effort required during incident investigation and help diagnosing production issues. This makes the paper directly relevant to AIOps, showcasing how LLMs can automate and assist with operational tasks in a DevOps environment.