Decentralized Agentic System Design

Thumbnail

Event details

Date 30.06.2025
Hour 09:0011:00
Speaker Sami Abuzakuk
Location
Category Conferences - Seminars
EDIC candidacy exam
Exam president: Prof. Bryan Ford
Thesis advisor: Prof. Anne-Marie Kermarrec
Co-examiner: Prof. Karl Aberer

Abstract
Recent systems for automated fault diagnosis increasingly
leverage large language models (LLMs) to support
or perform Root Cause Analysis (RCA). This report examines
three key papers that demonstrate how LLMs can be used to
structure, automate, and generalize RCA workflows. The first,
RCACopilot, presents a system that combines predefined diagnostic
workflows with LLM-based classification and summarization
for cloud incident analysis. The second, ReAct, introduces a
prompting framework that integrates reasoning and acting in
language models, enabling agents to alternate between internal
thought and external action. The third, RCAgent, extends this
framework by incorporating architectural mechanisms such as
expert delegation, structured tool interfaces, and trajectory selfconsistency
to support autonomous RCA. Together, these works
illustrate a progression from static RCA pipelines to dynamic
agentic systems. The report concludes with a discussion of open
challenges and outlines a research direction building on these
foundations.

Selected papers
  • Automatic Root Cause Analysis via Large Language Models for Cloud Incidents (link)
  • ReAct: Synergizing Reasoning and Acting in Language Models (link)
  • RCAgent: Cloud Root Cause Analysis by Autonomous Agents with Tool-Augmented Large Language Models (link)

Practical information

  • General public
  • Free

Contact

  • edic@epfl.ch

Tags

EDIC candidacy exam

Share