Resume

ZHANG Shuyuan - email - website

Education

• Imperial College London (2024 ~ 2025)
  • MSc Advanced Computing
• The University of Edinburgh (2020 ~ 2024)
  • BSc Computer Science (Honours)
  • Weighted average grade: 78 (First Class)

Highlights

Computer Vision / Computer Graphics

• UoE Course: Computer Graphics: Rendering & Computer Graphics: Geometry and Simulation
  • Learn C++ from scratch and implement Ray Tracing with advanced features including texture, acceleration hierarchy based on BVH, path tracing, pixel sampling, lens/aperture sampling and light sampling
  • Mesh Reconstruction from point clouds, discrete analysis and parameterization
• UoE Bachelor Dissertation: Inverse Procedural Modeling: from Sketches to Buildings
  • Construct a procedural model for buildings based on Directed Acyclic Graphs (DAG)
  • Distort 3D models and render as a 2D sketch image; generate training data with parameter sampling
  • Encoder-decoder & Multi-task decoders that predict DAG parameters based on input sketch image
  • Develop Blender add-on as user interface

Natural Language Processing

• UoE Coursess: Foundations of Natural Language Processing & Natural Language Understanding, Generation and Machine Translation
  • N-gram, Bayesian probabilities, RNN, GRU, LSTM, Transformer and Attention mechanism
  • Group Project for the course Machine Learning Practical: Query-focused Summarization via GPT Prompting on Ambiguous QA
• Extracurricular Application:
  • N-gram probabilities from Chinese poetry and as a name-generator mod for the game Stellaris, with over 500 subscriptions.

Large Language Models

• Application:
  • Prompt Engineering, Multi-agent systems, OpenAI API, Ollama API
  • Task breakdown for better performance and robustness via designing and implementing Multi-agent systems
  • Generate fine-tune data for low-resource tasks
  • Deployed on remote server QAnything to serve as a LLM backend with knowledge base

Digital Humans

• Deployed on remote server, extended and contributed to the open-source metahuman framework Linly_Talker, working with QAnything as an extra method to generate speech text for digital humans
• Deployed on remote server GPT-SoVITS to fine-tune models for voice cloning
• Deployed on remote server metahuman-stream as a real-time digital human streaming method

Internet of Things

• UoE Course: Principles and Design of IoT Systems
  • Collect human activity data through wearable devices
  • Design, implement and train neural networks (CNN, RNN, LSTM with model ensembling) to identify activity and respiratory symptoms
  • Develop Android App to deploy trained models, connect to sensors via Bluetooth and classify human activities /respiratory symptoms in real-time

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
  • Designed the domino-reloading and automatic placement mechanism
  • Designed and modelled the 3D printed parts
  • Wrote Python scripts to control the Lego motors
  • Coordinated Android App development
  • Designed and implemented the communication method between server, robot and the app

Work Experience

• HGTech - R&D Intern (Summer 2024)
  • Worked on digital human topics, deployed various services on compute server directly or using docker containers, hosted locally a Streamlit-driven webpage to track service status
• VisionTalk - Algorithms Intern (June 2024)
  • Improved LLM performance and robustness on chat generation and decision making via breaking down a complex task and designing/implementing a Multi-agent system
• GraphviX Lab, IPAB, UoE - Summer Research Internship (Summer 2023)
  • Researched on topics of sketch-based inverse procedural modeling. Implemented shape grammars based on Blender geometry nodes and its APIs. Trained neural networks to inference shape parameters from sketches, and integrated above pipelines as a Blender plug-in. Some tech stacks are also part of my bachelor dissertation.
• UoE - Tutor (2023)
  • A paid position delivering weekly tutorials for the course Reasoning and Agents, two hours per week
• EUFS - Software Infrastructure (2022 ~ 2023/24)
  • Software Infrastructure team for Driverless Vehicle
    • Maintained, refactored and extended eufs_cli, the command-line-interface tool in Python
      • refactored 70 lines of repeated code for each command to 5 lines
    • Provide supporting functionalities around git, colcon and more
    • Contributed to the server backend of EUFS-Testing-Application
    • Update launch configurations of the self-driving race car under different tasks
• Wuhan Tianyu Information Industry - Algorithms Intern (Summer 2021)
  • Carried out object detection tasks using OpenCV-Python and YOLOv5, trained models focusing on human and traffic analysis tasks. Wrote scripts to process large amount ofArUco codes. Collected, cleaned and augmented dataset for training