Hi, I'm Haotian Sun.
I'm a final-year Ph.D. candidate in Machine Learning at Georgia Institute of Technology, co-advised by Dr. Bo Dai and Dr. Chao Zhang. Most recently, I was a Student Researcher at ByteDance Seed in Summer 2026. Before that, I was an AI/ML Research Intern on Apple's Foundation Model Team, working with Dr. Nan Du, and an Applied Research Intern at Amazon, working with Dr. Liang Qiu.
I build agents that self-improve through interactions by co-designing efficient training algorithms, learnable harness frameworks, and scalable learning systems.
I've had the privilege of collaborating with researchers at
Apple,
ByteDance Seed,
Amazon,
and 
Georgia Tech.
News
- Jul 2026 Released STACX, an open-source modular infrastructure for end-to-end agentic RL.
- Jul 2026 Matryoshka Agent released on arXiv — unfolding sub-agents for long-horizon ML engineering.
- Jun 2026 Started as a Student Researcher at ByteDance Seed in San Jose, building a long-horizon multimodal data engine.
- Jun 2026 FReDA released on arXiv — forward-free diffusion language models with BPTT-free looped refinement.
- Dec 2025 Matryoshka Pilot appeared in the NeurIPS 2025 proceedings.
- Sep 2025 AmorLIP accepted to NeurIPS 2025.
- Sep 2025 Matryoshka Pilot accepted to NeurIPS 2025 (co-author).
- May 2025 Started my second research internship at Apple, Foundation Model Team, working on unified multi-modal diffusion.
- Jan 2025 EC-DIT accepted to ICLR 2025.
- Sep 2024 HYDRA accepted to NeurIPS 2024.
- Sep 2024 MedAdapter accepted to EMNLP 2024 (co-author).
- Sep 2024 Started as an Applied Research Intern at Amazon (Rufus) in Palo Alto, working on multimodal process supervision.
- May 2024 Started my first research internship at Apple, Foundation Model Team, working on scaling diffusion transformers.
- May 2024 BBox-Adapter accepted to ICML 2024 as a Spotlight (top 3.5%).
- Sep 2023 AdaPlanner accepted to NeurIPS 2023.
- Sep 2023 ToolQA accepted to NeurIPS 2023 (Datasets & Benchmarks Track).
Featured Projects
Modular Infrastructure for End-to-End Agentic RL
Changhao Li*, , Huijie Tang*, Rushi Qiang, Chenxiao Gao
Lab PI: Bo Dai
Open-source infrastructure 2026
Agentic RL, Training infrastructure, Sandboxed evaluation
Forward-Free Diffusion Language Models with BPTT-Free Looped Refinement
, Rushi Qiang, Yuqian Zheng, Bo Dai
arXiv preprint 2026
Diffusion language models, Forward-free training, Looped refinement
Unfolding Sub-Agents for Long-Horizon Machine Learning Engineering
Rushi Qiang, Changhao Li, , Yuchen Zhuang, Chao Zhang, Bo Dai
arXiv preprint 2026
Hierarchical agents, Long-horizon planning, ML engineering
Learning to Drive Black-Box LLMs with LLMs
Changhao Li, Yuchen Zhuang, Rushi Qiang, , Hanjun Dai, Chao Zhang, Bo Dai
Accepted at NeurIPS'25
Black-box LLMs, Agentic control, LLM-as-pilot
Efficient Language-Image Pretraining via Amortization
, Yitong Li, Yuchen Zhuang, Niao He, Hanjun Dai, Bo Dai
Accepted at NeurIPS'25
Vision-language pretraining, CLIP, Amortized inference
Scaling Diffusion Transformers with Adaptive Expert-Choice Routing
, Tao Lei, Bowen Zhang, Yanghao Li, Haoshuo Huang, Ruoming Pang, Bo Dai, Nan Du
Accepted at ICLR'25
MoE, Diffusion, Text-to-image synthesis
Model Factorization Framework for Black-Box LLM Personalization
Yuchen Zhuang, , Yue Yu, Rushi Qiang, Qifan Wang, Chao Zhang, Bo Dai
Accepted at NeurIPS'24
Black-box LLMs, Personalization, Recommendation
Lightweight Adapting for Black-Box Large Language Models
, Yuchen Zhuang*, Wei Wei, Chao Zhang, Bo Dai
Accepted at ICML'24 (Spotlight, top 3.5%)
Black-box LLMs, Adaptation, Plug-and-play
Adaptive Planning from Feedback with Language Models
, Yuchen Zhuang*, Lingkai Kong, Bo Dai, Chao Zhang
Accepted at NeurIPS'23
Closed-loop adaptive planning, LLM agent, Code interface, Skill discovery
A Dataset for LLM Question Answering with External Tools
Yuchen Zhuang, Yue Yu, Kuan Wang, , Chao Zhang
Accepted at NeurIPS'23 (Datasets and Benchmarks Track)
LLM evaluation, Dataset curation, Tool use