About the role#
As an Architecture Research Engineer Intern within the AGI Computing Lab, you will focus on the design challenges of serving Mixture-of-Experts (MoE) models at scale. You will work on distributed accelerators to improve performance and memory efficiency for future AI workloads.
What you'll need#
- Currently pursuing a master's or PhD in Computer Science or Electrical Engineering.
- At least one academic quarter or semester remaining in your program.
- Research experience in distributed ML systems or AI hardware like GPUs, TPUs, or NPUs.
- Experience developing performance models or simulators for large-scale ML systems.
- Understanding of system-level characteristics related to MoE inference and expert parallelism.
- Familiarity with PyTorch or JAX and LLM serving frameworks such as vLLM, SGLang, or TensorRT-LLM.
- Strong analytical, problem-solving, and communication skills.
- Ability to work both independently and as part of a team.
- Published papers in Microarchitecture, Systems, or ML Systems conferences are preferred.
- Background in optimization methods, such as graph partitioning or integer linear programming, is a plus.
What you'll do#
- Research expert placement, replication, and migration strategies for MoE models.
- Co-optimize expert placement with memory hierarchy, interconnect topology, and parallelism strategies.
- Analyze expert load imbalance, temporal drift, and cross-layer affinity using real-world workloads.
- Extend analytical simulators and design-space exploration tools to evaluate hardware design points.
- Collaborate with compiler and runtime teams to integrate placement-aware scheduling.
- Contribute to ML systems and architecture research, including writing research papers.
- Monitor advancements in LLM serving and ML architectures.
Location & details#
- This role is based in San Jose, California.
- The position requires daily on-site presence.
- This is a full-time, paid internship.

