Weijian Zhang
I build tools for open-weight models and software that agents can use
Right now I’m building Tuned Tensor — tools for defining how a model should behave and fine-tuning open-weight models to match.
Recent experiments: a local 2B email-triage model, an autonomous loop for prompt-injection classifiers, and Deeper Level, a chess interface that shows the model’s candidate moves and critique.
recent writing
| Aug 29, 2026 | MCP 2.0 |
|---|---|
| Jun 28, 2026 | Why I built a local 2B email triage model |
| Jun 13, 2026 | When Agent Loops Stop Moving |