AIGC Video Generation Algorithm (Leader)

Shopee; Singapore; On-site; SGD 180,000 - 260,000; Full time

Job summary

Shopee seeks a senior AI/ML leader to define the AIGC roadmap across pre- and post-training, including model alignment, data curation, and evaluation frameworks.

You will guide distributed training toolchains, optimize computation and storage, and drive production-ready inference with advanced distillation and quantization techniques. Leadership of a growing technical team is essential.

Qualifications

• Master's degree in a relevant field.
• 5+ years of AI/ML experience with a focus on generative models.
• Proven leadership of technical teams including hiring and mentoring.
• Hands-on experience in AIGC pre-training or post-training (or both).
• Deep familiarity with Transformer architectures and Diffusion models.

Responsibilities

• Define the AIGC roadmap across pre-training and post-training workstreams.
• Oversee distributed training toolchains and optimize training infrastructure.
• Lead post-training alignment, data curation, and RLHF/DPO/GRPO/PPO approaches.
• Drive capability expansion in long-video modeling and multi-modal generation.

Skills

Team leadership

AIGC expertise

Transformer architectures

Diffusion models

Distributed training

PyTorch

DeepSpeed

Megatron-LM

Communication

Education

Master's degree in Computer Science

Tools

PyTorch

DeepSpeed

Megatron-LM

Job description

Job Description

• Technical Strategy: Define the AIGC roadmap across pre‑training and post‑training workstreams — from distributed training infrastructure to alignment, evaluation, and inference deployment.
• Pre‑training & Infrastructure: Oversee the design of distributed training toolchains for ultra‑large‑scale AIGC models. Drive system‑level optimization across computation, communication, and storage layers. Ensure training stability and efficiency at scale.
• Post‑training & Alignment: Guide architecture design for video generation post‑training — including high‑quality instruction data curation, preference alignment (RLHF, DPO, GRPO, PPO), and video quality enhancement pipelines.
• Capability Expansion: Push the frontier on long‑video modeling, storyline consistency, precise camera control, and multi‑modal generation.
• Evaluation & Quality: Establish video quality evaluation frameworks and multi‑dimensional Reward Models to systematically measure and improve output quality.
• Inference & Deployment: Drive model distillation, quantization, and inference acceleration to bring research models into production.

Requirements

• Masters and above in Computer Science or any related field.
• At least 5 years of relevant experience in AI/ML, with a strong focus on generative models.
• Leadership: Demonstrated experience managing and growing a technical team with direct reports. Track record of hiring, mentoring, and retaining top talent.
• AIGC Depth: Hands‑on experience in AIGC pre‑training or post‑training (or both). Deep familiarity with Transformer architectures and Diffusion models (e.g., Stable Diffusion, Flux, DiT).
• Distributed Systems: Strong understanding of distributed training principles (Data/Pipeline/Tensor/Expert Parallelism) and frameworks such as PyTorch, DeepSpeed, and Megatron‑LM.
• Post‑training Expertise: Solid grasp of preference alignment methods (RLHF/DPO/GRPO/PPO), fine‑tuning techniques (LoRA/QLoRA/DoRA), and distillation approaches (Consistency Models, Flow Matching).
• Communication: Excellent cross‑functional communication skills. Comfortable presenting to senior leadership and collaborating across engineering, product, and research teams.

Plus Points

• Experience building a team or function from zero.
• End‑to‑end ownership of the full lifecycle of a video generation model, from data to deployment.
• Research leadership in physical simulation, world consistency, temporal consistency, or causal reasoning.
• Expertise in high‑quality video evaluation and human preference alignment at scale.
• Publications at top‑tier venues (NeurIPS, ICML, ICLR, CVPR, ACL, EMNLP).
• Familiarity with GPU hardware architecture, CUDA programming, NCCL, and cuDNN.
• Experience with extreme efficiency optimization such as inference acceleration, VRAM compression, quantization.