Student · Data Science · Portfolio

Diyuan Deng

B.Sc., Data Science and Big Data Technology (CUHK-Shenzhen) · Embodied AI, LLM Inference, Applied ML

About

Undergraduate focused on embodied AI / VLA data, enterprise LLM serving (SGLang, parallelism, quantization), interpretable clinical prediction, and full-stack engineering.

Education

The Chinese University of Hong Kong, Shenzhen

B.Sc., Data Science and Big Data Technology (expected 2027)

Sept. 2023 — Jun. 2027
  • GPA: 3.43 / 4.0
  • Coursework: Database Systems, Algorithm Design and Analysis, Stochastic Processes, Optimization, Machine Learning, Statistical Computing

Internship & Research

Data Engineering Intern · Shanghai Wuzhi Evolution Technology

Embodied Data R&D · Shanghai

2025.06 — 2025.08

Collaborated with Prof. Bo Zhao's group (SJTU) on embodied AI data and VLA models.

  • Built the xArm data pipeline and optimized VLA; 10,000+ high-quality trajectories.
  • Adapted VLA to LeRobot; multimodal LLMs and natural-language (NL) command interfaces.
  • Grasping and multi-layer stacking tasks: success rate above 90%.

Research Assistant · NSC-Advisor (Ischemic Stroke Prognosis)

Clinical prediction & prototype UI

2025.09 — 2025.12

Related work submitted to ISMRM (under review).

  • Interpretable favorable vs. unfavorable prognosis from admission data.
  • Stratified sampling; few-shot and prompting; RAG for conflicting clinical evidence.
  • Dual-input UI and visualization; 93.75% on 16 held-out real cases.

Projects

SF Technology · Enterprise LLM inference (SFGlang)

Core Developer

Jan. 2026 — Present

  • Throughput and communication for Qwen3-class LLMs and multimodal models (4B / 30B) under enterprise load.
  • SGLang and dual-container setup on RTX 4090 / H20 clusters; PD/EPD, TP/DP/EP, FP8/INT8; load-test web platform.
  • Qwen-30B TP4+DP2: TTFT −48.9%, +664% throughput; Qwen-4B TP1+DP2: +750% throughput.

Library Management System

Developer

Sept.–Nov. 2024

  • Full-stack Python (tkinter) + SQLite; normalized schema and composite indexes.
  • 15+ features including SHA-256 auth and admin dashboard; tests covered all requirements.

2D RPG combat (UE5 + C++)

Developer

Sept.–Dec. 2024

  • C++ combat with three enemy AI types; Qt launcher with SHA-256 messaging; PaperZD animation state machines.
  • Team core codebase 1,000+ lines; best course project.

Campus Roles & Community

Undergraduate Teaching Fellow (USTF)

Sept. 2024 — Present

School of Data Science

Data Structures, Databases, and one more core course; 10+ tutorials for 800+ students; co-designed problems and grading; SDS Excellent USTF Award (2024–2025).

Student Assistant, University Library

Nov. 2023 — Present

University Library

Printers (Bambu Lab X1E, Ultimaker S5, Formlabs Form 2), 30+ jobs; Final Cut Pro and Adobe Photoshop edits; 3D-printing workshops.

2Tired Cycling Club

Jul. 2024 — Dec. 2025

Founder & Vice President

Core team of 20 across five functions; Songshan Lake and Tour de France meet-ups; 80+ members in year one; ~95% satisfaction.

Honors

  • CUMCM Guangdong Third Prize (2025)
  • Bronze, men's 1500 m (University Games)
  • Bronze, men's 100 m breaststroke (Sports Festival)

Skills & Languages

Technical

Languages

Chinese (native); English (professional proficiency for coursework and work).

Interests

Cycling, photography, hiking, amateur radio, Model United Nations.

Contact

Open to internships, research collaboration, and technical discussions.