AI Agent / LLM Infrastructure Engineer building production agent systems across model routing, tool execution, cloud capacity, observability, and autonomous workflows.
I focus on systems where agents do real work: run tools, call models, manage cloud capacity, publish artifacts, and leave traces that can be debugged.
| Project | What It Proves | Stack |
|---|---|---|
| ai-shorts-matrix-pipeline | AI Shorts matrix pipeline: script, storyboard, AI video, FFmpeg polish, batch upload, multi-channel scheduling, and YPP-oriented cost tracking | Python, Typer, FFmpeg, Langfuse, YouTube Data API |
| agentic-chrome-fuzzing-harness | Agent-driven security testing harness: LLM-generated fuzz inputs, ASan feedback, tmux workers, crash capture, and dashboard | Chromium, ASan, Codex/Claude-style agents, Bash, Python |
| jobclaw | AI-powered job search agent: scrape jobs, match against a profile, draft applications, notify, and track outcomes | AI agents, Playwright, LLMs, automation |
| titan-builder-mcp | MCP server that lets AI agents interact with Ethereum builder / MEV infrastructure through tool calls | Rust, MCP, Ethereum |
- Agent pipelines: multi-stage workflows with retries, review points, typed artifacts, and traceable model calls.
- LLM infrastructure: model routing, quota/capacity planning, GPU/TPM operations, observability, and cost control.
- Automation tools: browser automation, publishing automation, job-search automation, and creator workflow tooling.
- Security-adjacent agent systems: long-running fuzzing/security-test harnesses and vulnerability intelligence workflows.
I also track fast-moving Chinese open-source AI projects and package English starter guides:
DeepSeek V4 hot track
- deepseek-awesome-integration-v4-companion — Companion checklist for the official DeepSeek integration index.
- deepseek-v3-to-v4-migration-notes — Migration notes from DeepSeek V3 to V4.
- vllm-deepseek-v4-serving-notes — vLLM serving notes for DeepSeek V4.
- llamafactory-deepseek-v4-finetune-notes — LLaMA-Factory fine-tuning notes for DeepSeek V4.
- new-api-deepseek-v4-routing-guide — new-api routing guide for DeepSeek V4 Pro/Flash.
- deepseek-v4-claude-code-bridge — Claude Code-style workflow bridge for DeepSeek V4.
- deepseek-v4-opencode-starter — OpenCode starter for DeepSeek V4 Pro agentic coding.
- deepseek-v4-openai-sdk-starter — OpenAI SDK migration starter for DeepSeek V4.
- deepseek-v4-anthropic-sdk-starter — Anthropic-compatible API starter for DeepSeek V4.
- deepseek-v4-long-context-rag — 1M-context RAG evaluation guide.
- deepseek-v4-agent-bench-kit — Agentic coding and tool-use benchmark kit.
- deepseek-v4-cost-calculator — Pro vs Flash cost planning starter.
- deepseek-v4-tool-calling-lab — Tool calling and JSON action reliability lab.
- deepseek-v4-fim-coding-starter — Fill-in-the-middle coding workflow starter.
- deepseek-v4-prompt-evals — Prompt evaluation starter for V4 Pro and Flash.
GPT Image 2 hot track
- youmind-gpt-image-2-companion — Companion workflow for YouMind GPT Image 2 prompt galleries.
- gpt-image-2-prompts-api-companion — API companion for GPT Image 2 prompt collections.
- go-openai-gpt-image-2-notes — go-openai notes for GPT Image 2 backend workflows.
- gpt-image-2-skill-companion — Agent skill companion for GPT Image 2 generation/editing.
- gpt-image-playground-2-companion — Playground companion for reproducible GPT Image 2 experiments.
Agent frameworks and runtime
- qwenpaw-english-starter — QwenPaw English field guide.
- qwen-agent-cookbook-en — Qwen-Agent cookbook for MCP, RAG, function calling, and code interpreter workflows.
- agentscope-english-guide — AgentScope English field guide for production multi-agent systems.
- agentscope-runtime-english-guide — AgentScope Runtime guide for deployment, sandboxing, and observability.
- spring-ai-alibaba-english-guide — Spring AI Alibaba guide for Java enterprise AI workflows.
- chatdev-english-guide — ChatDev guide for multi-agent software engineering experiments.
Multimodal, speech, and retrieval
- minicpm-o-live-starter — MiniCPM-o live multimodal agent starter.
- cosyvoice-english-guide — CosyVoice guide for text-to-speech and voice generation workflows.
- funasr-english-guide — FunASR guide for speech recognition and real-time ASR workflows.
- flagembedding-english-guide — FlagEmbedding guide for BGE embeddings, retrieval, reranking, and RAG.
- internvl-english-guide — InternVL guide for multimodal vision-language model evaluation.
LLM deployment and training
- lmdeploy-english-guide — LMDeploy guide for LLM inference, serving, and deployment.
- xtuner-english-guide — XTuner guide for LLM fine-tuning and training workflows.
- Packaging public proof repos around agent infrastructure and creator automation.
- Turning private production work into clean open-source examples.
- Writing technical notes on Claude Code, Codex, MCP, model routing, and practical agent operations.
- Portfolio: https://zijiezhong.com
- LinkedIn: https://linkedin.com/in/j-z-57327b2b5