B.S. Electrical Engineering, National Taiwan University
Built my foundation in ML, NLP, computer vision, speech, and data science.
Trustworthy AI · ML systems · Agentic AI
Building AI you can actually trust to run on its own.
I'm Tsung-Huan, an MSCS student at Georgia Tech, originally from Taiwan. Previously, I spent two years at Academia Sinica working across the data, training, and evaluation stack for LLM/VLM systems: VLM data pipelines and pre-training, LLM safety post-training, safety evaluator data, and adversarial attacks.
At Georgia Tech, I'm studying the stack behind AI systems that are capable, measurable, and deployable. Machine learning, deep learning, and deep reinforcement learning sharpen the modeling layer; data visualization and analysis and database concept & design ground the data layer; and hardware-software co-design for ML keeps me thinking about performance when models leave the notebook. In summer 2026, I'm joining LinkedIn to work on AI agent platforms.
Off the record: workouts, One Piece, and a playlist that can't decide between Chinese pop and R&B.
Timeline
Scroll ↓ to advanceResearch assistant at Academia Sinica
Worked across VLM data and pre-training, LLM safety post-training, evaluator data, and adversarial attacks.
NeurIPS Safe Generative AI Workshop
Presented work on preserving safety in fine-tuned LLMs.
MSCS at Georgia Tech
Broadening into ML infrastructure, deep learning, RL, data systems, and HW-SW co-design.
AI/ML Intern at LinkedIn
Joining LinkedIn in summer 2026 to work on AI agent platforms.
Projects
Scroll ↓ to advance
course-project VisionMealRL
An end-to-end meal-planning class project that connects visual nutrition estimation with a DQN planner, recommending meals and portions across multi-day horizons while balancing nutrition goals, user preference, and diversity.
Deep RLcomputer visionnutrition AI
research GenAI Safety Assessment
Safety and security evaluation for Taiwanese LLMs, covering safeguard modeling and automatic red-teaming.
Generative AILLM safetyred-teaming
ongoing Taiwan Visual-Language Model 70B
Continual pre-training of ViT-L + Llama3-Taiwan-70B on 50M images and 3B text tokens grounded in Taiwanese culture.
Generative AImultimodallarge-scale pre-training
course-project Speech Bot
A Jetson Nano robot combining lightweight ASR, OpenAI conversation API, TTS, and hand pose detection for multimodal interaction.
speech processingedge device