AI Researcher (LLM / MoE / MHA / Tokenization)

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

Job summary

MS Capital Singapore seeks talented researchers to advance MoE and large-model capabilities. You will work on mixture of experts, tokenization, and transformer-based systems using PyTorch/JAX, with emphasis on scalable training and memory optimization.

The team combines researchers and engineers to push the frontier of AI-driven trading tech, under a strong AI R&D charter in Singapore.

Qualifications

• Master or PhD in Computer Science, AI, Mathematics, or related fields.
• Research in MoE/MHA/Tokenization with publications or in-depth projects.
• Proficient in PyTorch or JAX, with experience in large model training and fine-tuning.
• Deep understanding of Transformer architectures with exposure to Megatron/DeepSpeed.
• Familiarity with distributed training and memory optimization.
• Strong academic curiosity plus engineering mindset for practical implementation.

Skills

MoE/MHA/Tokenization research

PyTorch/JAX

Transformer architectures

Distributed training

Engineering mindset

Self-driven exploration

Education

Master or PhD in Computer Science / Artificial Intelligence / Mathematics

Tools

Megatron

DeepSpeed

Job description

MS Capital is a private fund management company with a strong founding team with long-accumulated experience in strategy modelling, trading system and platform development. Using advanced artificial intelligence technology as the cornerstone, and enforcing strict investment management, the company's investment fund has gained sustained and stable returns.

You will be joining MS Capital's technology arm, with AL/ML as its cornerstone, and is committed to providing users with high-quality and stable trading services. The company now has a number of experienced quantitative researchers, world-class deep learning scientists and engineers from leading internet companies and top universities. The company has also provided various kinds of trading solutions for a number of leading brokerage firms and organizations. The company's vision is to integrate artificial intelligence technology with quantitative investment scenarios, relying on strong artificial intelligence R&D capabilities and advanced trading strategy models, to provide users with comprehensive and stable investment service.

We are dedicated to breakthrough research in the core technologies of large models and are looking for talented individuals with research experience and strong interest in following areas:

Mixture of Experts (MoE)

• Sparse activation mechanisms, dynamic routing algorithms, load balancing optimization
• Efficient training/inference architectures for ultra-large-scale MoE
• Integration and innovation between MoE and traditional Transformers
• Efficient attention mechanisms (e.g., sparse attention, linear attention)
• Theoretical analysis and performance improvement of attention structures
• Design of multimodal and cross-modal attention mechanisms
• Next-generation tokenization algorithms (subword/character/byte-level optimization)
• Quantitative impact analysis of tokenization on model performance
• Unified frameworks for multilingual/cross-lingual tokenization
• Embedding compression and semantic space optimization

Skills & Qualifications

• Master / PhD in Computer Science, Artificial Intelligence, Mathematics, or related fields;
• Research experiences in MoE/MHA/Tokenization, with published papers (preferably in top conferences) or in-depth research projects.
• Proficient in frameworks such as PyTorch/JAX, with experience in large model training and fine-tuning;
• Deep understanding of Transformer architectures, with experience for related source code (e.g., Megatron, DeepSpeed);
• Familiarity with distributed training and memory optimization is a plus;
• Strong academic curiosity + Engineering mindset for practical implementation + Self-driven exploration.