QA Lead: Agentic AI & Network Testing
Posted Updated
Description
Zinkworks is seeking a hands-on QA Lead to own quality for Network Digital Twin and agentic AI use cases in telecom and broadband. Define how we test AI agents that reason, call tools and act on live data, and build the testing framework and evaluation harnesses.
You will lead QA strategy, write test scenarios, and work with AI engineers, developers and architects on cloud-native systems running on GCP. Strong background in software testing and agentic testing is required.
Define the QA strategy for agentic software systems. Own end-to-end QA for agentic apps, including agents, tools, prompts, and RAG pipelines. Define quality gates for non-deterministic systems and establish acceptance criteria. Set offline evaluation, pre-release regression, and production monitoring of agent behavior. Validate Digital Twin fidelity against real network and use as a safe sandbox for testing. Build and run testing, and maintain evaluation harnesses and golden datasets. Test REST and event-driven microservices interactions for correctness and permissions. Design adversarial and safety tests: prompt injections, jailbreaks, data leakage, guardrail bypass. Integrate automated tests into CI/CD for backend and agent interactions. Lead QA across the team and mentor engineers on AI testing practices. Track quality metrics and regressions to leadership. Ensure agents are testable by design with observability hook points. Participate in agile ceremonies and drive testing discipline. 7+ years in software testing with team leadership experience. Experience testing distributed microservices, REST APIs and event-driven systems. Hands-on testing of LLM-based or agentic applications. Knowledge of evaluation methods: golden datasets, LLM-as-judge, trajectory and tool-call evaluation. Experience with evaluation/observability frameworks such as LangSmith, Langfuse, DeepEval, promptfoo or Ragas. Strong Python (Pytest) for test automation and CI/CD integration. Understanding AI risks: hallucination, prompt injection, data leakage and unsafe actions.Skills
Sign in to read and join the discussion.