SOTA！Community
I worked with AI researchers and early LLM developers to turn emerging models, papers, and engineering experiments into practical knowledge the community could understand, test, and build on.
Early developers and shared practice
I created Open Source Daily and Developer LIVE to discover developers experimenting with open-source models, fine-tuning, and Agents. Together, we documented their technical choices, implementation process, and lessons learned, turning individual experiments into knowledge others could reuse.
Developer LIVE
Conversations with developers about real engineering practices.
Building an Agent from 0 to 1 · 2023
Worked with an early Agent developer to document Llama 2 fine-tuning on 6×RTX 4090, instruction decomposition, function calling, task planning, tool execution, and deployment.
ProjectFrom research labs to developers
I helped produce 25+ sessions with AI labs and original research teams, including M6, GLM-130B, and Alpaca-CoT. I organized technical conversations and turned high-value questions into concise notes on research methods, technical decisions, and practical boundaries.
Testing models through real tasks
I created 15+ hands-on evaluations of open-source models, including DeepSeek-Coder-V2, Mistral, and Qwen. Instead of repeating benchmark scores, I tested coding, reasoning, and tool-use tasks to help developers understand how the models behaved in practice.