Multimodal Vision Research Intern (Agentic AI)

Tencent; Singapore; On-site; SGD 120,000 - 180,000; Full time

Job summary

Tencent in Singapore seeks a researcher to advance Agentic Vision for Hunyuan multimodal foundation models. You will research visual reasoning post-training, integrating perception, planning, tool use, and environment feedback to enable dynamic, interactive reasoning with images and video.

You will develop multimodal post-training methods (SFT, PPO/GRPO, OPD), build verifiable data synthesis pipelines, and create interactive sandboxes for code execution and visual tools.

Qualifications

• Master's degree or higher in Computer Science, Artificial Intelligence, Computer Vision, Machine Learning, Mathematical Statistics, or a related field.
• Strong foundations in deep learning and algorithm implementation, proficiency in PyTorch, and experience with distributed training.
• Experience with multimodal foundation models, visual reasoning, multimodal agents, or reinforcement learning for post-training, plus familiarity with SFT, PPO/GRPO, OPD, RLVR, or reward modeling.

Responsibilities

• Conduct research and development on Agentic Vision for Hunyuan multimodal foundation models, focusing on visual reasoning post-training.
• Advance visual reasoning and spatial intelligence across fine-grained perception, 3D understanding, and complex visual logic.
• Develop multimodal post-training methods (SFT, PPO/GRPO, OPD) and verifiers/reward models for long-horizon tasks.
• Build verifiable data synthesis pipelines using rendering engines, symbolic systems, and code sandboxes.
• Create interactive sandboxes supporting code execution, visual tools, and evaluation methods.

Skills

PyTorch proficiency

Distributed training

Multimodal reasoning

Reinforcement learning

Education

Master's degree or higher in Computer Science/AI/ML/Statistics

Tools

Megatron

OpenRLHF

Job description

Tencent in Singapore seeks a researcher to advance Agentic Vision for Hunyuan multimodal foundation models. You will research visual reasoning post-training, integrating perception, planning, tool use, and environment feedback to enable dynamic, interactive reasoning with images and video.

You will develop multimodal post-training methods (SFT, PPO/GRPO, OPD), build verifiable data synthesis pipelines, and create interactive sandboxes for code execution and visual tools.