EurekaClaw
Lead development of the EurekaClaw project: designed the frontend interface, coordinated open-source community contributions, and maintained the codebase for research and public use.
Projects
Web World Models (WWM)
A framework that generates unlimited context via LLMs in text/code environments without data storage. Collaboration with Princeton AI Lab.
TENG++: Boundary-Enhanced PINNs
Extended the Time-Evolving Natural Gradient framework with Dirichlet boundary constraints in Physics-Informed Neural Networks.
Predicting Text from Intracranial Neural Signals
CNN-Transformer decoder with RoPE embeddings for intracortical speech-to-text, achieving CER reduction from 0.234 to 0.1636.
Deep RL Algorithms in Cryptocurrency Trading
Off-policy actor-critic agents (DDPG, TD3, SAC) for ETH trading, with SAC achieving best returns.