LLM serving · RAG · agent systems
Building & operating production AI stacks · Wuxi, China
I ship and operate production LLM / RAG / agent systems — self-hosted Dify ops, retrieval pipelines, and upstream bugfixes in active AI repos that interviewers recognize (Dify, RAGFlow, Qwen Code, Alibaba MNN, ModelScope ms-swift, QwenPaw, and related tooling).
Prefer fixes with reproducible RCA + tests, not drive-by docs.
Upstream contributions · LLM serving · RAG · agent runtimes
vLLM · Dify · RAGFlow · Qwen · MNN · ms-swift · Milvus · AgentScope
| Area | Role |
|---|---|
| langgenius/dify | Contributor — API / Web / Agent runtime bugfixes |
| infiniflow/ragflow | Contributor — Agent retrieval / Wiki indexing (×2 merges) |
| QwenLM/qwen-code | Contributor — serve boot / MCP discovery timing |
| alibaba/MNN | Contributor — converter fail-fast / TFLite index + createUnit (×2 merges) |
| modelscope/ms-swift | Contributor — distributed eval / generate path |
| agentscope-ai/QwenPaw | Contributor — tool-call coordinator reliability |
More work in flight across Qwen / agent / RAG ecosystems.
Thanks for stopping by — happy to chat about LLM infra, RAG, and agent systems.