AI Evaluation Engineer

This position is listed on behalf of a partner company, which manages all applications and next steps. Our partner is looking for an AI Evaluation Engineer based in Romania. As a founding AI Evaluation Engineer, you will design and build the evaluation, testing, and observability infrastructure that ensures production-grade AI agents perform reliably, safely, and efficiently. You will develop automated testing systems for complex reasoning, multi-turn conversations, and agent interactions involving APIs, databases, and code execution. Working across AI engineering, software testing, and infrastructure, you will create scalable evaluation pipelines that go beyond traditional QA and manual prompt testing. You will establish robust regression testing, statistical evaluation methods, and adversarial scenarios to identify weaknesses before deployment. This is an opportunity to take ownership of an evaluation architecture from the ground up while working with modern AI technologies in a fully remote, globally distributed environment. Your work will directly contribute to the reliability, security, and continuous improvement of advanced AI systems. Accountabilities - Design, build, and maintain automated CI/CD testing pipelines to validate AI agent behavior before deployment and throughout the development lifecycle. - Develop and maintain golden datasets containing regression prompts, edge cases, and synthetic scenarios to measure agent accuracy, safety, performance, and latency. - Implement fast, cost-effective, deterministic evaluation checks before introducing LLM-as-a-judge approaches, including JSON schema validation, regular expressions, API status checks, and embedding similarity. - Create comprehensive test suites for multi-turn agent conversations, evaluating context retention, state consistency, reasoning behavior, and safeguards against infinite tool-calling loops.

- Design and implement adversarial testing scenarios to assess resilience against prompt injection, malformed inputs, unexpected tool responses, and other potential security vulnerabilities. - Build statistical evaluation frameworks to measure non-deterministic AI behavior using pass rates across multiple runs, confidence intervals, and other quantitative reliability metrics. - Integrate evaluation workflows into GitHub Actions or comparable CI systems using SDK-level execution and automated testing practices. - Implement observability and tracing solutions using tools such as OpenTelemetry, LangSmith, Arize Phoenix, W&B Weave, or equivalent platforms to monitor agent execution and diagnose failures. - Optimize evaluation efficiency and cost through caching, mocking, and other techniques that reduce unnecessary LLM API calls. - Validate structured outputs, including JSON schemas, OpenAPI specifications, and function-calling payloads, to ensure agents interact reliably with external systems. - Evaluate and improve agent performance under load, accounting for variable response times and non-deterministic endpoint behavior. - Independently design and own the evaluation architecture, establishing best practices, testing standards, and scalable processes for ongoing quality assurance. Requirements - Strong proficiency in Python or TypeScript, particularly for API testing, headless execution, asynchronous programming, and automated testing systems. - Hands-on experience building LLM evaluation pipelines, AI testing frameworks, or statistical testing systems capable of measuring performance across repeated runs. - Experience integrating automated tests into CI/CD environments, particularly GitHub Actions or similar platforms, with SDK-level execution. - Strong expertise in validating structured data and outputs, including JSON schema, OpenAPI specifications, and function-calling payloads.