Research Kernel Engineer

About Bitdeer

Bitdeer is a leading global technology company specializing in Bitcoin mining and AI cloud solutions. They provide end-to-end Bitcoin mining services including ASIC chip design, mining rig manufacturing, equipment procurement, logistics, data center design and operation, and network management. With headquarters in Singapore and a global presence across multiple countries, Bitdeer operates a diversified 3 GW energy portfolio and runs mining and HPC data centers worldwide.

About Bitdeer AI Lab

Bitdeer AI Lab is a cutting-edge division within Bitdeer focused on pioneering artificial intelligence research with the aim of making AI inference affordable and accessible. The lab invests in infrastructure and software innovations that optimize AI model deployment and performance.

Role Overview

The Research Kernel Engineer will accelerate the lab's research by developing efficient computational kernels. Their work will involve transforming novel AI methods like unique quantization approaches, sparsity patterns, and attention mechanisms into highly optimized CUDA and Triton kernel implementations tailored for specific models and workloads. They will analyze performance bottlenecks through advanced profiling and ensure that algorithms which lack efficient open-source implementations become production-ready.

Key Responsibilities

• Design and optimize CUDA and Triton kernels for novel AI research methods developed by the team.
• Conduct detailed performance analysis including roofline modeling to identify computational hotspots under real traffic conditions.
• Transform slow reference implementations of published research into production-level, efficient code.
• Develop reliable measurement frameworks to support ongoing research and optimization efforts.
• Collaborate closely with AI Cloud platform engineers responsible for the serving infrastructure.

Qualifications and Skills

• Bachelor's, Master's, or PhD in Computer Science, Electrical Engineering, or a similar technical discipline with practical experience in GPU programming.
• Proven expertise in writing and tuning CUDA and/or Triton kernels along with strong proficiency in Python and C++.
• Deep understanding of GPU architectures, memory hierarchies, occupancy optimization, memory coalescing, tensor cores, and warp-level primitives, supported by profiling tools such as Nsight Compute or equivalent.
• Experience optimizing inference-critical operations like attention, GEMM, normalization, sampling, and KV-cache management.
• Able to implement novel methods based on research papers or informal prototypes without existing reference code.
• Strong ability to design meaningful metrics prior to optimization and critically assess performance results.
• Experience with inference engines such as vLLM, SGLang, or TensorRT-LLM, especially in writing custom kernels or extensions, is highly advantageous.
• Familiarity with compiler or intermediate representation frameworks like MLIR, TVM, or TorchInductor, or experience with distributed serving parallelism is a plus.
• Publications in leading systems conferences or significant open-source contributions in GPU computing or inference projects are welcomed.
• Passion for advancing AI infrastructure, hardware performance optimization, high ownership, and disciplined software engineering practice.

Work Environment and Benefits

• A culture that embraces authenticity, diversity, and open-minded dialogue.
• An inclusive, respectful workspace with a startup-like energetic atmosphere.
• Opportunities to connect with industry leaders and innovators as the company scales rapidly.
• Direct impact on the future of digital asset technology through involvement in pioneering projects.
• Participation in defining and improving internal processes and systems.
• Highly autonomous role encouraging personal accountability and fast professional growth.
• Comprehensive benefits package including welfare perks, training, and mentorship programs.