About
I am a PhD student in the Department of Computer Science and Technology at the University of Cambridge, supervised by Dr. Rika Antonova and Prof. Robert Mullins, and part of CamRAL. I previously worked with Dr. Edward Johns in the Robot Learning Lab at Imperial College London.
My PhD is funded by AIxSIM, an ARIA-funded cross-stack simulation effort for novel AI hardware, where I also work closely with Dr. Aaron Zhao, Prof. Timothy M. Jones and Dr. Jianyi Cheng. I build hardware for agents, and agents for hardware.
I want to see a compute stack that rebuilds itself, where agents become the new paradigm of design space exploration, redesigning the very models, kernels, compilers and instruction sets they run on, with every step checked by proof rather than trust.
Research, up the stack
- Self-improving agents
Agents that evolve alongside their own evaluators: Red Queen Gödel Machine
- Serving
Splitting inference stages across specialised hardware: When Does Disaggregation Pay?
- Kernels
Close-to-metal kernel generation for emerging hardware, kernel optimisation for GPUs, and formal verification of kernels: KernelCraft, AReaL-TIK, LARK
- Accelerators
NPU quantisation and design space exploration for agentic inference:PLENA, MemExplorer, diffusion language models
- Microarchitecture
Design space exploration for branch predictors: PIP
Education
-
-
Master's in Computing (AI and Machine Learning), Imperial College London
-
Bachelor's in Computer Science, University of Manchester
Industry
-
Microsoft Research
-
Amazon
-
Arm
Teaching
-
Teaching Assistant, University of Cambridge
-
PASS2 Leader, University of Manchester
Publications * Equal contribution
Talks
Service
© 2026 Jiayi Nie