ServiceNow projects
ServiceNow / INCIDENT INTELLIGENCEProduction experience

Multi-Agent Orchestration for Autonomous Root-Cause Analysis

Retrieval and agent workflows that bring incident history into investigation and triage.

PythonLangChainCrewAIVector databases
SYSTEM OVERVIEW
the pieces, connected.
Current incident (Investigation context) + Incident history (Related events and records) → Domain retrieval (Relevant historical context) → Agent orchestration (LangChain + CrewAI) → Historical patterns (Similar incidents) + Alert prioritization (Investigation workflow) → Root-cause hypotheses (Context for investigation)

A public sketch of the documented components and workflow. Internal interfaces are omitted.

Download full-resolution SVG
THE PROBLEM

Where the work started.

Incident investigation depends on context scattered across historical records. Teams need to find related events, identify patterns and form useful root-cause hypotheses.

MY CONTRIBUTION

What I built.

  • Engineered multi-agent workflows for incident investigation and context gathering.
  • Built a domain-specific RAG pipeline over historical incident information.
  • Used retrieval and agent orchestration to support alert prioritization, pattern identification and root-cause hypotheses.
ENGINEERING APPROACH

How the system comes together.

01

Retrieve incident context

Domain-specific RAG

Built retrieval over historical incident information to bring relevant prior events into root-cause analysis.

02

Coordinate investigation

LangChain · CrewAI

Engineered incident-management agents and orchestration for investigation and alert prioritization.

03

Support a useful next step

Historical patterns · root-cause hypotheses

Connected incident history with agent workflows to identify patterns and produce context for triage and root-cause investigation.

THE RESULT
60%less manual triage time

Supported incident investigation with historical context and automated parts of the manual triage workflow.

Selected work from my engineering role at ServiceNow. Impact figures are reported from this work; the companion examples below use synthetic data.

CODE & EVALUATION

Inspect the work behind the explanation.

SYNTHETIC COMPANION EXAMPLE

Classification accuracy, coverage and abstentions

A new, runnable example that scores saved predictions against expected outputs. It uses fictional data and illustrates evaluation mechanics; it does not reproduce the production system or substantiate the impact figures above.

python3 evaluate_outputs.py incident-triage.json report.json

Python 3.10+ · Standard library only · Includes fixtures, metric definitions and tests

Explore more projects Discuss this project