I build agent-driven automation — content operations, research distillation, and turning repeated work into reusable skills.
| Project | What it is |
|---|---|
| llm-building-map | An agent skill on how LLMs are actually built — data pipeline, post-training, evaluation — with every claim mapped to a timestamp in the source lecture. MIT. |
| solana_coralOS | Starter kit for agents that earn: sellers bid over Coral Protocol / MCP, settlement goes through a Solana escrow. |
| qingtong-training-report | Automated completion report for construction-site safety training. |
- Hackathon / bounty work: open-innovation-solana, imperial-agent-economy, polish-solana-research
- Writing: superteam-talent-portal-article
- If it can be automated, it doesn't get done by hand.
- If it can be verified, it isn't accepted as "looks right" — tests, screenshots, read-back.
- Deliverables stay self-contained: no external services, runs offline, licence permits closed-source commercial use.
我用 AI agent 做自动化：内容运营、资料提炼，以及把重复劳动变成可复用的技能。
llm-building-map 是我最近开源的一个技能包：把"大模型是怎么造出来的"讲清楚（数据管线、后训练、评测），每条结论都能对回原课的时间码，可以直接装进 Claude Code 或 Hermes 用。
做东西的习惯：能自动化的不手工，能验证的不靠"看起来对"，交付物做到自包含、离线可用、许可允许商用。
📫 有问题直接开 issue，比自己猜快。