AI TRIAGE AGENT: REAL BUG OR FALSE FAILURE?

Internship

• Field: AI ENGINEERING
• Duration: 6 months

Problem

Many nightly test failures are not real product bugs, including locator changes, slow pages, popups, used data, or unavailable environments. Each failure is replayed and triaged manually, causing real bugs to be reported late.

Key tasks

• Build an AI agent that replays failed test steps on the application by label and role rather than broken locators.
• Turn vague expected results into concrete UI, network, console, and database checks.
• Classify each failure as a real bug, script issue, environment issue, data issue, or flaky failure.
• Draft bugs in Azure DevOps.
• Enforce safety through test-environment-only execution, an action allow-list, a step budget, and an auditable trace.

Success criteria

• Verdict accuracy of at least 80% compared with QA decisions.
• Zero actions outside the allowed scope.
• Less triage time per failure.
• False failures closed without manual replay.

Candidate profile

• Final-year engineering student in software, AI, or data.
• Python, LLM APIs, prompt and agent design.
• Playwright or Selenium and Azure DevOps API are a plus.
• Curious, research-minded, and comfortable with uncertain outputs.