Résumé

Leyan Li

Performance engineering for scientific computing, and agent systems built on what those measurements produce.

Availability

Coursework and dissertation complete 2026.08
Available to start 2026.09.01
Degree certificate issued 2026.12

Coursework and the dissertation both finish in August 2026, so I can work full time from September.

Education

University of Edinburgh

MSc High Performance Computing with Data Science

University of Liverpool

BSc Computer Science, First Class Honours.

Technical skills

claims the projects can back
Programming & engineering Python, C/C++, SQL · Linux, Bash, Git, pytest
Agent & LLM applications Agent architectures, prompt engineering, function calling, eval design
Backend & databases FastAPI, SQLite/MySQL, REST/HTTP fundamentals
ML & high performance computing PyTorch, pandas, NumPy · MPI, OpenMP, scaling analysis

Projects

full detail →

Agent-style performance analysis over HPC benchmark data

  • Imported real benchmark runs into SQLite behind deterministic analysis tools.
  • Added an LLM function-calling router scored on the same eval set: 83.3% vs 100%, ablation 63.3%.
  • Hardened the boundary so failures are explicit rather than guessed.
  • Kept it checkable: deterministic reports, structured tracing, 229 offline tests.

github.com/leyancode/hpc-benchmark-agent

Value factorisation in multi-agent actor-critic methods

  • A single-variable comparison of three critic designs inside MAPPO.
  • Result against intuition: the simplest critic won.
  • Traced the cause and wrote the negative results into the dissertation.

github.com/leyancode/marl-value-factorisation

PETSc solver scaling study on ARCHER2

  • Ran a controlled MPI+OpenMP layout study up to 32 nodes / 4,096 cores.
  • Hybrid beat flat MPI in both dimensions; 29.83× at 93.2% efficiency in 2D.
  • Retracted an earlier stage after an audit; 408 audited runs carry the conclusions.
  • Located the mechanism with Linaro MAP profiling.

github.com/leyancode/hpc_benchmark_archer2

Research & publication

Convolutional Neural Networks Based Medical Image Analysis

Li, Leyan. EMITI 2024.

String Art Generator Based on the Radon Transform

An image-to-string-art pipeline with preprocessing and a parameter interface.

Contact

A PDF résumé is available on request.