Ongoing Projects

Autonomous Skill Learning for LLM Agents

Feb 2026 - Present

Building a self-evolving agent framework that learns transferable skills from execution experience with demonstrated transfer across terminal, web, and structured data environments.

• Designed a domain-agnostic memory architecture supporting skill generalization beyond code tasks
• Implemented HPC-compatible execution for large-scale multi-domain evaluation
• Demonstrated successful cross-domain skill transfer in diverse environments

Automated AI-Researcher Agent with Trend-Aware Idea Generation

Jan 2026 - Present

Developing an autonomous system that identifies high-impact research ideas from literature trends and produces publication-ready manuscripts end-to-end.

• Architected multi-dimensional evaluation (novelty, feasibility, impact) to filter ideas before costly experimentation
• Building end-to-end pipeline from idea generation to manuscript writing

Compositional Generalization of World Models

May 2025 - Present

Designing a prototype-tree-based framework for extracting & reusing atomic features from training data to generalize in unseen scenarios at inference time.

• Finetuning a unified world model with object & action prototype injection
• Enhancing both video generation quality and simulation task performance