AI Research Engineer (Video Vision)

Background: Our client is an immersive media tech company selected for a large-scale government-led R&D national project in a consortium with domestic conglomerates and national research institutes. They are developing Asia's first 8K-class ultra-large LED immersive content space, alongside a proprietary AI video engine that expands and converts existing films and videos to fit this spatial specification. The goal of this engine is to reduce production costs by over 60% compared to traditional CG/VFX methods and significantly shorten production timelines. Amidst the immersive experience market growing at an annual average rate of 20%, they possess a roadmap starting with the establishment of the first domestic space and expanding into major Asian cities and the global market within the next 10 years. The company's AI development team values research that rigorously verifies whether a problem is truly solved to the end, rather than merely quickly applying the latest papers and models. It is a team that asks: Why is it necessary? Is it genuinely better than existing methods? How will it be measured? Can it run in our GPU environment? Can it be applied to actual services? The Video / Vision AI Research Engineer we are seeking is a position that leads core research for this AI video processing platform. You will directly define and solve challenging problems in AI processing for movie-length, long-duration videos, such as Video Generation, Outpainting / Inpainting, high-resolution video processing, and Temporal Consistency. Research results do not end in papers or prototypes; they proceed to actual production through SOTA exploration, reproduction, dataset construction, training and fine-tuning, evaluation, optimization, and checkpoint management. This provides an opportunity to experience research turning directly into commercial services, and to shape the technical direction of the research organization alongside business expansion. Key Responsibilities: Research, investigate, and validate latest papers, models, and open source in Video Generation / Video Processing; Research and advance Video Outpainting / Inpainting models; Improve Temporal Consistency and Scene Consistency in long-duration videos; Research technologies to maintain Identity / Object Consistency for people and objects; Research Style Consistency to preserve the mood and expression of original videos; Research video expansion and conversion technologies for dome and ultra-wide content generation; Research and optimize high-resolution video processing and 12K/16K processing pipelines; Implement and reproduce AI model papers and open source; Collect, preprocess, refine, and build training data pipelines for datasets; Build training / fine-tuning / inference pipelines; Design benchmarks and evaluation metrics for model performance evaluation; Profile and optimize model performance including GPU memory, throughput, and latency.