Lead AI Video Generation Model Evaluation Engineer

Video Rebirth Limited; Hong Kong; On-site; HKD 900,000 - 1,300,000; Full time

Job summary

Video Rebirth Limited is seeking an experienced Head of Evaluation Engineering to own the 0-to-1 strategic roadmap for global model evaluation. You will design and scale an industrial-grade, distributed evaluation infrastructure to provide rapid feedback for massive training runs.

You will lead Proprietary Metric R&D, benchmark governance, and HITL/RLAIF data pipelines, translating artistic and cinematic quality standards into measurable signals and working closely with founding researchers.

Qualifications

• Bachelor’s degree or above in a relevant field.
• Proven ability to drive cross-functional collaboration.
• Experience in video-generation model evaluation and benchmarking.
• Proficient in prompt engineering and narrative structuring.

Responsibilities

• Own the 0-to-1 strategic roadmap for global model evaluation.
• Design and scale distributed evaluation infrastructure for rapid feedback during massive training runs.
• Lead research and implementation of next-generation automated metrics for multi-modal evaluation.
• Architect scalable HITL and AI Feedback pipelines (RLAIF).
• Partner with founders and leads to diagnose model regressions and guide a world-class team.

Skills

Technical leadership

Prompt engineering

Cross-functional collaboration

AI/evaluation

Communication

Education

Bachelor’s degree or above

Job description

- Technical Vision & Roadmap: Own the 0-to-1 strategic roadmap for our global model evaluation strategy; design and scale an industrial-grade, distributed evaluation infrastructure to provide rapid feedback for massive core training runs.

- Proprietary Metric R&D: Lead the research and implementation of next-generation automated metrics that outpace industry baselines (solving the limitations of FVD/CLIP); pioneer custom frameworks for evaluating multi-modal alignment, fluid physics, and spatiotemporal continuity.

- Benchmark Engineering & Governance: Establish the company's definitive evaluation benchmarks, curating high-difficulty datasets that stress-test complex camera dynamics, extreme motion controls, multi-character consistency, and cinematic aesthetics.

- Data Operations & Alignment Loops: Architect highly efficient, scalable Human-in-the-loop (HITL) and RLAIF (AI Feedback) pipelines; translate subjective artistic/cinematic standards into quantifiable data signals to drive direct model alignment.

- Cross-Functional Leadership: Partner closely with Founding Researchers and Core Training Leads to diagnose model regressions (e.g., temporal flickering, morphing, semantic drift); build, mentor, and technically lead a world-class evaluation engineering team as the startup scales.

Qualifications & Requirements

• Requirements / Qualifications
• Education: Bachelor’s degree or above.
• Core Competencies: Strong analytical, communication, and fast-learning skills with a proven ability to drive effective cross-functional collaboration.
• Domain Experience: Prior industry experience in video generation model evaluation and benchmarking.
• Prompt Engineering: Proficient in Prompt Engineering, with a deep understanding of how to craft prompts that effectively articulate narrative structure, physical commonsense, and motion dynamics/tension.
• Product Mindset: Strong product sense with the ability to objectively evaluate model outputs from diverse user perspectives.
• Visual & Cinematic Knowledge: Solid background in fine arts, visual design, film production, or photography, with a clear understanding of camera language, depth of field, and framing.
• Visual Aesthetics & Detail Sensitivity: High sensitivity to color and image details, capable of evaluating generated outputs through composition, perspective, lighting, color theory, and fine grain details.