Python / AI developer. I currently build AI agent applications at Jiangxi Yunyan Shijie, and previously led backend development of an AI platform and LLM application orchestration system at Mengchuang Shuangyang. I focus on shipping LLM applications, AI agent architecture, and model deployment on Ascend NPUs.
📫 Email: zhu05361@gmail.com
- Languages & frameworks: Python, FastAPI, asyncio, Pydantic, LangChain, pytest
- LLMs: LLM deployment & optimization, Embedding / Rerank, SFT fine-tuning, RAG, vLLM / vllm-ascend
- Agents: Orchestrator–worker agents, async sub-agents, MCP / FastMCP tool integration
- Engineering & infrastructure: Linux, Docker, Git, Redis, MySQL, Ascend NPU / CANN
- Bid-writing agent: An agent that covers material analysis, writing plans, chapter generation, and Word document delivery. v0.1 is live with real users.
- LLM application orchestration platform: A config-driven FastAPI backend with parallel DAG workflow execution, a plugin system, dynamic MCP tools, SSE streaming, and conversation memory.
- Jinboshi AI Assistant (Q&A module): Deployed Embedding / Rerank models on Ascend NPUs and trained an intent classifier on Qwen2.5-0.5B, which is 32.1% faster and 21.4% more accurate than the previous qwen-max approach.
- vllm-ascend-setup: Automation scripts and a troubleshooting guide for setting up a vLLM development environment on Ascend NPUs.
- husk: A lightweight agent sandbox server.
- [InternLM-tutorial] Wrote course docs and served as a teaching assistant for the InternLM LLM bootcamp (PR #253)
- [CodeCamp2023-488] Added new configuration files for the SoftTeacher algorithm in mmdetection (PR #10856)
- [MMSIG] Tested and adapted MMDet's new config system (PR #11113)
- [MMSIG] Added the ovdet algorithm to mmdet projects (PR #11217)
- Hubei University of Automotive Technology, M.Eng. in Electronic Information (Computer Technology), 2022 – 2025
- Hubei University Zhixing College, B.Eng. in Computer Science, 2018 – 2022
- Vaasa University of Applied Sciences, exchange student in Information Engineering, 2019