Job Details:
Job Description:
We are building a next-generation LLM inference system, spanning model optimization, inference runtime, and system-level design.
This is a research + engineering role where you will:
- Study cutting-edge work (LLM inference, MoE, system optimization)
- Implement and optimize core techniques
- Work across the stack: model → kernels → runtime → distributed system
A key focus is GPU kernel and runtime optimization, including exploring Triton-like programming models and compiler approaches, as part of building an end-to-end AI rack software system for LLM inference.
Key Responsibilities
1. Research & Prototyping
- Read and reproduce state-of-the-art work (LLM inference, MoE, systems)
- Translate ideas into working, optimized implementations
- Identify bottlenecks and iterate beyond baseline performance
2. LLM Inference Optimization
Implement and evaluate techniques such as:
- Continuous / dynamic batching
- KV cache optimization and memory management
- Speculative decoding
- Flash / paged attention
- Quantization (INT8 / FP8 / low-bit)
Optimize for latency, throughput, and GPU utilization
3. MoE (Mixture-of-Experts) Systems
Explore efficient inference for sparse models:
- Routing strategies and load balancing
- Expert parallelism and sharding
- Communication vs computation trade-offs
Improve scalability and efficiency of MoE inference
4. Kernel & Runtime Optimization
Develop and optimize GPU kernels using modern approaches:
- Triton-like programming models
- CUDA or equivalent low-level frameworks
Investigate:
- Memory access patterns and layout optimization
- Operator fusion and kernel efficiency
- Compiler-style optimization for tensor workloads
Compare different kernel/runtime strategies and integrate into the system
5. End-to-End Inference System Development
Build and optimize a full inference stack:
- Model execution layer (vLLM, TensorRT-LLM, or similar)
- Runtime scheduling and batching
- Distributed inference across GPUs/nodes
Work on:
- Multi-GPU / multi-node scaling
- NCCL / communication optimization
- System-level performance tuning
Qualifications:
Basic Requirements
- Master’s or PhD student required (CS, EE, or related field)
- Strong programming skills (Python required)
- Familiar with PyTorch and transformer models
- Solid fundamentals in algorithms and systems
- Available for at least 6 months (shorter durations are not considered)
Preferred Experience
Experience with one or more:
- GPU programming (CUDA, Triton, or similar)
- LLM inference frameworks (vLLM, TensorRT-LLM, FasterTransformer)
- Distributed systems or parallel computing
Knowledge of:
- GPU architecture and performance profiling
- Quantization or model optimization
- MoE or large-scale model systems
What We Look For
- Ability to go from paper → implementation → optimization
- Strong interest in performance and system-level problems
- Fast execution and willingness to work on deep technical challenges
- Curiosity about how LLM systems actually run at scale
Job Type:
Student / Intern
Shift:
Shift 1 (China)
Primary Location:
PRC, Shanghai
Additional Locations:
PRC, Beijing, PRC, Shenzhen
Business group:
The Sales and Marketing Group (SMG) leverages the product portfolio to drive Intel's revenue growth and market expansion, blending strategic initiatives with dynamic sales efforts to capture and retain customers. SMG is responsible for empowering the sales force with tools and insights needed to close deals and build lasting customer relationships. Sales analytics and market research ensure strategies are both targeted and impactful. In SMG, disciplined execution, creativity, and ambition are celebrated, providing ample opportunities for career advancement and skill development.
Posting Statement:
All qualified applicants will receive consideration for employment without regard to race, color, religion, religious creed, sex, national origin, ancestry, age, physical or mental disability, medical condition, genetic information, military and veteran status, marital status, pregnancy, gender, gender expression, gender identity, sexual orientation, or any other characteristic protected by local law, regulation, or ordinance.
Position of Trust
N/A
Work Model for this Role
This role will require an on-site presence. * Job posting details (such as work model, location or time type) are subject to change.
- ADDITIONAL INFORMATION: Intel is committed to Responsible Business Alliance (RBA) compliance and ethical hiring practices. We do not charge any fees during our hiring process. Candidates should never be required to pay recruitment fees, medical examination fees, or any other charges as a condition of employment. If you are asked to pay any fees during our hiring process, please report this immediately to your recruiter.