About the role#
The AGI Computing Lab focuses on system-level challenges created by the rising demands of AI and ML workloads. This internship involves researching agentic benchmarks like SWE-bench and WebArena to understand their performance characteristics on distributed memory accelerator servers. You will contribute to architecture research and help build models that inform the design of future accelerators.
What you'll need#
- Currently pursuing a master's or PhD in Computer Science or Electrical Engineering.
- Research experience in workload characterization or performance analysis of CPUs, GPUs, and NPUs.
- Experience with profiling tools such as Nsight Systems or PyTorch Profiler.
- Experience developing simulators.
- Familiarity with PyTorch, agent frameworks, and LLM serving frameworks like vLLM or SGLang.
- Strong analytical, problem-solving, and communication skills.
- Ability to work both independently and as part of a team.
- Published papers in microarchitecture, systems, or characterization venues are preferred.
What you'll do#
- Research system-level characteristics of agentic AI benchmarks.
- Develop end-to-end profiling of agents across LLM inference, tool calls, and environment execution.
- Characterize memory and communication demands of long, multi-turn contexts and KV-cache reuse.
- Build workload models to drive analytical simulators and design-space exploration tools for future accelerators.
- Collaborate with compiler and runtime teams on hardware and software co-design optimizations.
- Write research papers and stay current with trends in agentic AI and ML architectures.
Location & details#
- This is an on-site internship based in San Jose, California.
- The position is full-time and paid.
- This role does not offer visa sponsorship.

