- BCI startup Algorithm Engineer → transitioning to LLM Algorithm Engineer
- Building LLM Inference Infrastructure: custom operators, serving optimization, vLLM internals
- Designed & implemented Multi-Agent Systems with production-grade context engineering
- Researching RL Post-Training Infrastructure: environment orchestration, observability, non-stationarity
|
|
|
|
|
|
Building a Personal Super-Agent System — an interconnected fleet of multi-agent projects covering the full lifecycle of information.
| Acquire / 信息获取 | Process & Accumulate / 信息处理与知识沉淀 | Train / RL 训练基础设施 | Output / 内容输出 |
|---|---|---|---|
| Sailor 水手 |
Maelstrom 大漩涡 |
Forge 锻造厂 |
Coming Soon |
| Multi-Agent RSS Crawling LLM Scoring & Clustering |
Multi-Agent Doc Processing Context Engineering & RAG Long-term Knowledge Store |
Controllable Environments Training Observability Single-Node RL Training |
LLM Content Generation Multi-channel Distribution |
|
Sailor | Information Acquisition Engine A sailor navigating diverse seas of information. Multi-agent RSS aggregation system with LLM-powered analysis — 5-stage preprocessing pipeline, AI article scoring & clustering, and automated knowledge base insight reports. FastAPI React LLM Agent RSS SQLite
|
Maelstrom | Information Processing & Knowledge Hub A maelstrom that devours the flood of documents and distills structured knowledge. Multi-agent document processing pipeline with context engineering — translation, review, terminology & QA agents collaborating on bilingual Markdown output, with RAG-powered long-term knowledge accumulation and retrieval. Next.js FastAPI Multi-Agent PaddleOCR PyMuPDF
|