OpenThoughts-Agent: Data Recipes for Agentic Models
2026
Negin Raoof, Richard Zhuang, Marianna Nezhurina, ... Michael Siu, and 24 more
arXiv available soon
AI Researcher / Software Engineer / Builder
I study how language models learn, improve, and act in realistic environments.
I'm pursuing a B.S. in Computer Science at the University of Southern California ✌️
My research focuses on post-training, eval, and self-improving agents. I'm currently a technical staff intern at Kimi (Moonshot AI), and I work with Yue Zhao and Jieyu Zhao at USC.
Reach out to me at siuw [at] usc [dot] edu
Research Publications
AD-AGENT: A Multi-agent Framework for End-to-end Anomaly Detection
2025
Tiankai Yang*, Junjun Liu*, Michael Siu*, Jiahang Wang, Zhuangzhuang Qian, Chanjuan Song, Cheng Cheng, Xiyang Hu, Yue Zhao
Findings of IJCNLP-AACL, 2025
Pyod 2: A python library for outlier detection with llm-powered model selection
2025
Sihan Chen*, Zhuangzhuang Qian*, Michael Siu*, and 10 more
The Web Conference 2025, Demo Track
* Equal contribution
Selected Projects
AD-AGENT
Multi-agent framework for automating anomaly detection across different data modalities using LLMs.
Monolith
Infrastructure for recursive language models, enabling AI coding agents to reason persistently across arbitrarily large contexts without context-window rot or lossy summarization.
OpenThoughts-Agent
Open-source datasets for training agent-based LLMs.
DineOn
A mobile app for USC students to check daily dining hall menus and food locations.
Beyond Research
I listen to a lot of R&B, jazz, and hip-hop, and I've played basketball for years. The way I think about rhythm and structure carries over to how I build things.
“There is no if's, would's, could's, should's; it just is, and we just are.”