I am a third-year Computer Science + AI undergraduate at HKUST, deeply passionate about human-centered intelligent systems — AI that truly understands people and helps them live and work better.
My experience spans multimodal learning (video, audio, language, physiological signals & sensors), pre-training & fine-tuning, model deployment, intelligent system architecture, evaluation frameworks, and rapid vibe-coding prototyping.
My goal is to build practical, human-aligned AI that bridges perception, cognition, and action to create meaningful real-world impact.
StateGuard — Ongoing Personal Project
GitHub → https://github.com/L-Ark/StateGuard
(Supervised by Prof. Ma Xiaojuan, HKUST)
Lightweight, privacy-preserving, real-time cognitive and emotional state monitoring system for knowledge workers.
- Gated multimodal architecture fusing rPPG (heart rate & HRV) and facial Valence-Arousal from webcam.
- Closed-loop Just-in-Time Adaptive Interventions with calm technology design.
- Evolved from Tsinghua HCI+ summer research on face foundation models.
Multi-person Re-Identification via Cross-Modal Skeleton–IMU Alignment — Ongoing Research Project
Best Project Award — HKUST Embedded AI Systems Course
(Supervised by Prof. Ouyang Xiaomin, HKUST)
End-to-end pipeline for long-term multi-person identity maintenance by fusing camera skeleton tracks with wearable IMU data (targeted at autism early screening).
- Developed hybrid synthetic IMU generation and history-based greedy matching strategies.
- Tackled data scarcity and accumulated error in cross-modal alignment.
- Project Report
Leveraging Diffusion Models for Music Source Separation — EPFL VITA Lab Exchange
Project Report
Waveform-native generative approach for high-fidelity music source separation (drums, bass, guitar, piano).
- Integrated Stable Audio Open autoencoder with Scalable Interpolant Transformer (SiT) using Flow Matching.
- Achieved leading structural fidelity (SSIM 0.59) among generative models.
Self-Supervised Face Foundation Model (HCI+ 2025)
Tsinghua Pervasive HCI Lab Collaboration
Mainly responsible for video-based facial emotion prediction and recognition.
- Integrated physiological signals (rPPG) and eye-tracking features into a self-supervised face foundation model.
LLM-Hanabi — Paper
Co-first author @ EMNLP Workshop 2025 (Wordplay)
Designed and implemented a multi-agent benchmark for evaluating Theory of Mind and cooperative behaviors in LLMs using the game of Hanabi.
- Built the full environment, ToM metrics, automated testing pipeline, and conducted comprehensive experiments.
eBPF-LLM NetSentinel — EPFL CS-477 Advanced Operating Systems Course
GitHub → https://github.com/NagatoBigSeven/eBPF-LLM-NetSentinel
Intelligent real-time network threat detection and blocking system.
- Combines eBPF XDP kernel-level high-speed packet filtering with LLM-powered threat analysis (multi-backend support).
- Features Human-in-the-Loop validation, Streamlit dashboard, whitelist management, and proactive IP banning.
-
Research Intern — MiniMax (Hailuo AI)
Evaluated LLM factuality and hallucination benchmarks. -
Summer Research — Tsinghua Pervasive HCI Lab (HCI+ 2025)
Face foundation models for emotion recognition and physiological monitoring. -
Project Leader — HKUST RoboMaster Enterprise Robot Team (2024)
Led mechanical design and system integration for Launcher-Dart subsystem.
Programming & Development
- Languages: Python (Advanced), C++, Linux/Unix
- Frameworks & Tools: PyTorch, Hugging Face, OpenAI API, eBPF, Streamlit, Git (advanced version control & collaboration)
Mathematics & Foundations
- Multivariate Calculus, Linear Algebra, Probability & Statistics, Discrete Mathematics, Algorithms, Optimization Methods
Artificial Intelligence
- Machine Learning, Deep Learning, Embedded AI Systems
- Specialized: Multimodal Learning & Fusion, Video & Audio Processing (VideoMAE, Diffusion Models), rPPG, Facial Emotion Recognition, LLM Integration & Evaluation, Multi-Agent Systems
Mechanical Design & Robotics
- SolidWorks, AutoCAD, Bamboo Studio, CNC Machining, 3D Printing, Prototyping & Assembly
Others
- Operating Systems, Software Construction, Model Deployment, System Architecture
- HKSAR Government Scholarship — Talent Development Scholarship
- HKUST Undergraduate Scholarship — University’s Scholarship Scheme for Continuing Undergraduate Students
- Dean’s List Academic Achievement Medal (Top ~3%, 2023–2025)
- Best Project Award — HKUST Embedded AI Systems (2026)
- RoboMaster 2024: 1st Prize (Regional) & 2nd Prize (Final)
- EMNLP Workshop 2025 Publication
- Email: fliangae@connect.ust.hk
- GitHub: @L-Ark
- Phone: +86 13590552186 / +825 59335964
Actively seeking research internships or collaborations in Human-Centered AI, Multimodal Intelligence, Affective Computing, and Embodied Systems (2026).