ZHANG Shuyuan

A dedicated Computer Science graduate from the University of Edinburgh, currently pursuing an MSc in Advanced Computing at Imperial College London. With a solid background in Computer Vision, Computer Graphics, and Natural Language Processing, he has gained valuable experience in various technical domains, including digital humans and large language models.

News

Want bashrc-level customization of your Windows Terminal?

• check this note

Teaser on my ISO project (multi-agent training-free shape program synthesis): (yes, cherry-picked results)

TL;DR

CV/CG

• UoE Bachelor Dissertation: Inverse Procedural Modeling: from Sketches to Buildings
  • Blender add-on, DAG, Distortion Renderer, Encoder-decoder, Multi-task decoders
• took all (both) CG courses available at UoE:
  • C++ Ray Tracer, Mesh Reconstruction from point clouds, Discrete Analysis and Parameterization

NLP

• UoE NLP Coursess: Foundations of Natural Language Processing & Natural Language Understanding, Generation and Machine Translation
  • from N-gram, Bayesian probabilities to RNN, GRU, LSTM and finally Transformer and Attention mechanism
• NLP for games: built Stellaris namelist mod based on Ngram probabilities computed in a Chinese poetry dataset (500+ subs!)

LLMs

• no money so focused on applications:
  • Prompt Engineering, Multi-agent systems, OpenAI API, Ollama API
  • Generate fine-tune data for low-resource tasks

Digital Humans

• Deployed / Integrated open-source frameworks and made contributions:
  • Linly_Talker
  • GPT-SoVITS
  • metahuman-stream
  • QAnything

Internet of Things

• UoE Course: Principles and Design of IoT Systems
  • wearable devices, human activity recognition, deploy trained models to android

System Engineering

• Group Project: A domino-placing robot based on TurtleBot, Lego motors, 3D printed parts and an Android App for Bluetooth connection to the robot for the course System Design Project

Software Engineering

• Backend: flask, fastapi, MongoDB
• Dockerize services, CLI based off argparse / simple Web UI with streamlit