I'm an M.S. student in Computational Science and Engineering at Harvard University, cross-registered at MIT. I work on multimodal model post-training, long-horizon agent systems, video understanding, and scalable machine learning systems.
Previously, I was a Machine Learning Engineer Intern at TikTok, working on multimodal AI and agent systems, and at Baidu, working on multimodal and speech AI.
- Amphion — Contributor and co-author of the Amphion v0.2 technical report for open-source audio, music, and speech generation.
- Debatts — First-author zero-shot debating TTS system with style-conditioned, emotionally adaptive speech interaction.
- Implicit Chain-of-Vision — First-author multimodal reasoning framework; CVPR 2025 MAR Workshop Oral.
- VideoResearcher — Active-perception and self-evolving tool-use agents for long-video understanding.
Multimodal LLMs · Post-training · Agent Systems · Video Understanding · Reinforcement Learning · ML Systems
- MemFlow: Learning Compact Soft Memory for Long-Context Personalization — Co-first author, ICLR 2027 submission. OpenReview
- MLP Cache: Trust-Region Computation Reuse — Second author, ICLR 2027 submission. OpenReview
- Autonomous Multimodal Reasoning via Implicit Chain-of-Vision — First author, CVPR 2025 MAR Workshop Oral. Paper
- E3-System: Expressive Emodata, Explicitly Controllable, Easy-to-Use TTS — First author, IEEE ISCSLP 2026.
- Overview of the Amphion Toolkit (v0.2) — Co-author. Paper
- Debatts: Zero-Shot Debating Text-to-Speech Synthesis — First author. Paper · Dataset
Python · PyTorch · Hugging Face · VLM Post-training · GRPO · Agent Tool Use · Megatron-LM · DeepSpeed · vLLM · Distributed Training