Agent Algorithm Evaluation Engineer

• Evaluation Framework Design: Design and build multi-dimensional evaluation metric systems tailored to diverse business scenarios.
• Red-Team Testing & Safety Control: Simulate complex, extreme real-world user scenarios to conduct stress testing and red-teaming of AI models; identify and drive remediation of issues related to bias, hallucinations, policy-violating content, and values alignment.
• In-Depth Bad Case Analysis: Conduct root-cause analysis of erroneous model outputs and collaborate with algorithm engineers (Model/SFT/RLHF) to drive prompt optimization and fine-tuning improvements.
• Automated Evaluation Tooling: Leverage LLM-as-a-Judge methodologies to build automated evaluation pipelines that enhance product iteration efficiency.
• User Experience Insights: Conduct in-depth research into user psychology within social and companion-style products, translating subjective, qualitative experience into quantifiable metrics to continuously enhance product experience.

• Proven experience enhancing automated evaluation workflows, including but not limited to conventional evaluation methods, LLM-as-a-Judge, and automated weakness mining.
• Proven experience enhancing automated data production pipelines, including but not limited to Self-Instruct, Evol-Instruct, and Synthetic Data RL.
• Strong software engineering fundamentals, with proficiency in Python and Shell scripting and hands-on experience in Linux environments.
• Master's degree or above in Computer Science or a related field, with at least 3 years' working experience.

Nice-to-Have

• Prior background in AI evaluation, QA automation, or NLP algorithm development at a major internet company.
• Extensive experience as a power user or developer of mainstream AI social products.