Hi there! I currently focus on building multi-agent systems to generate synthetic data for LLM post-training and alignment.
I am happy to chat and discuss potential collaborations, feel free to reach out by
🌟 Studying Zone
I am collaborating with Cornell ICPC and Millennium to build efficient LLMs for code and data generation.
- This work is called ALICE (Aligning Language models for Interactive Code Execution), find more about it at alicellm.github.io.
- ALICE is a meta-agent collaboration system that generates high-quality data through multi-turn interactions and feedback without human intervention.
- It produces multimodal data with traces from agent strategies like ReAct and Reflexion, which are scarce but offer potential for aligning advanced LLMs.
Previously, I led the prior work of ALICE called Voice2Action with Cornell XRC, an Unity Package for real-time code execution in VR.
I am also working on large-scale generation augmented retrieval systems (opposed to RAG) at Cornell NLP.
I used to work on graph machine learning at AWS AI Lab (2021-2022) and contribute to the open source Deep Graph Library.
👀 Chilling Zone
I like programming! I lead the "Cornell Tech" Group at Cornell ICPC and won the Top 20% in 2023 Regional!
I enjoy cooking, listening to music of all forms, playing ping-pong, reading science fiction, and more!
⚡ Developing Zone
📈 "Accepted" Zone