Yingwei Noah Song
Research Engineer · PhD Student · Solo Developer
San Francisco, California
I’m a Ph.D. student in Electrical & Computer Engineering at the University of Arizona, where I’m a Herbold Fellow advised by Prof. Huanrui Yang, and a founding research engineer at a stealth AI startup. I work on video generation, 3D-grounded world models, and the agents that tie them together. I care about the whole stack: training the model, building the infrastructure around it, and shipping the product people actually use.
Education
University of Arizona
Ph.D. in Electrical & Computer Engineering · Herbold Fellow
Advisor: Prof. Huanrui Yang
Brown University
M.S. in Computer Science
GPA 4.00 / 4.00
University of Wisconsin–Madison
B.S. in Computer Science & Data Science, minor in Math
GPA 3.90 / 4.00
Latest Work
Stealth Startup
Founding Research Engineer · Industry
Seed-stage AI startup, 10M+ USD raised. Technical lead owning over 70% of the company’s engineering across model research, AI infrastructure, backend, frontend, and product design.
Hedra
Research Engineering Intern · Industry
Contributor to Character-3 (MagicInfinite), Hedra’s flagship omnimodal foundation model for digital-human video — launched to 3M+ users and anchoring Hedra’s a16z-led $32M Series A.
Harvard Visual Computing Lab
Research Assistant · Research
Worked with Prof. Hanspeter Pfister and Dr. Wanhua Li on 4D LangSplat, bringing language fields into dynamic 3D Gaussian splatting (CVPR 2025).
Selected Projects
Stealth
AI-UGC marketing video SaaS
Advisor and principal developer. A SaaS that generates AI-UGC marketing videos and social-media marketing strategy from a product link.
3D-Grounded World Model
Exploratory research
Lead. Preliminary research on a new world-model architecture: a first bridge between 3D models and video models.
Palette
Short-drama generation agent
An AI agent that generates complete short dramas end-to-end.
AgentShot
Agentic multi-shot video generation & benchmarking
A 5-stage plan-then-generate pipeline of 9 specialized LLM/VLM agents with tool use.
Publications
LIVEditor-14B: Lightning Unified Video Editing via In-Context Sparse Attention
ICML 2026Accepted
AgentShot: Towards Agentic Multi-Shot Video Generation and Benchmarking
AAAI 2027Under reviewHVCap: Human-centric Video Captioning via Factorized GRPO and a New Benchmark
ACM MM 2026Under review
4D LangSplat: 4D Language Gaussian Splatting via Multimodal Large Language Models
CVPR 2025Accepted
VisitorsWhere people read this site from
- Visits
- —
- Countries
- —
- Cities
- —