The Problem

Enterprise automation systems drift silently. A workflow that passed testing last sprint may behave differently today due to API changes, data schema shifts, UI updates, or model behavior changes — and nobody notices until production breaks.

Traditional test suites catch known failures. Driftex catches unknown ones.


Inspiration

Most enterprise QA teams spend 60–70% of their time maintaining existing tests, not writing new ones. The root cause: automation brittleness caused by behavioral drift. We wanted to build an agent that monitors, detects, and explains drift — autonomously.


What Driftex Does

Driftex is an AI-powered testing agent built on UiPath Test Cloud that:

  • Monitors enterprise automation workflows continuously
  • Detects behavioral drift by comparing current execution patterns against historical baselines
  • Diagnoses root causes using LLM-powered AI analysis
  • Alerts human testers with actionable fix recommendations
  • Self-heals simple test breakages automatically

How We Built It

  • UiPath Test Cloud — core test execution and orchestration
  • UiPath Agent Builder — drift detection agent logic
  • UiPath Maestro — orchestrating the full agentic workflow
  • Claude Code — used to build and refine automation scripts
  • Python — anomaly detection and baseline comparison logic

Architecture Flow:

  1. Test Cloud runs scheduled test executions
  2. Results feed into the Drift Detection Agent
  3. Agent compares results against historical baseline
  4. LLM analyzes anomalies and generates root cause diagnosis
  5. Human-in-the-loop alert sent for critical drifts
  6. Auto-remediation triggered for known drift patterns

What We Learned

  • UiPath Maestro is powerful for orchestrating multi-agent flows
  • LLMs are surprisingly effective at explaining why a test failed, not just that it failed
  • Human-in-the-loop design is critical for enterprise trust

Challenges

  • Defining what "drift" means quantitatively across different workflow types
  • Balancing auto-remediation vs. human oversight
  • Integrating external LLM analysis within UiPath's orchestration layer cleanly

Built With

  • agent
  • builder
  • claude
  • cloud
  • maestro
  • python
  • test
  • uipath
Share this project:

Updates

Submission history