Publications

TrioXpert: An Automated Incident Management Framework for Microservice System

Yongqian Sun1,7, Yu Luo1, Xidao Wen4,*, Yuan Yuan6, Xiaohui Nie2, Shenglin Zhang1,5, Tong Liu3, Xi Luo3

1Nankai University   2Computer Network Information Center, Chinese Academy of Sciences   3Lenovo (Tianjin) Co., Ltd.   4BizSeer
5Key Laboratory of Data and Intelligent System Security, Ministry of Education, China   6National University of Defense Technology   7Tianjin Key Laboratory of Software Experience and Human Computer Interaction

ASE 2025

* Corresponding author

Paper Code Slide Story
TrioXpert teaser figure
TrioXpert integrates multimodal observability data with collaborative LLM-based experts for automated incident management.

Abstract

TrioXpert is an end-to-end incident management framework for microservice systems. It jointly considers anomaly detection, failure triage, and root cause localization, and combines multimodal observability signals with LLM-based collaborative reasoning. Instead of treating metrics, logs, and traces as a single undifferentiated input stream, TrioXpert extracts structured evidence from each modality and lets specialized experts reason over the evidence in an interpretable workflow.

Project Overview

Multimodal Evidence

Metrics, logs, and traces are processed according to their own data characteristics, turning raw observability streams into evidence that can support diagnosis.

Collaborative Experts

LLM-based experts cooperate across incident management tasks, making the diagnosis process more transparent than a single black-box prediction.

End-to-End Management

The framework covers anomaly detection, failure triage, and root cause localization in a unified workflow for microservice incidents.

Method

TrioXpert method overview
The method overview highlights how evidence from multiple modalities flows into collaborative incident management experts.

Citation

@inproceedings{sun2025trioxpert,
  title={TrioXpert: An Automated Incident Management Framework for Microservice System},
  author={Sun, Yongqian and Luo, Yu and Wen, Xidao and Yuan, Yuan and Nie, Xiaohui and Zhang, Shenglin and Liu, Tong and Luo, Xi},
  booktitle={Proceedings of the 40th IEEE/ACM International Conference on Automated Software Engineering},
  year={2025}
}