A Chinese roadmap for programmers entering LLM engineering

What it does

llm-master is a Chinese-language knowledge base that arranges LLM engineering topics into a six-stage progression: model basics, API application, RAG, Agent design, production deployment, and Transformer internals. Each stage defines concrete deliverables—such as a RAG system with offline evaluation benchmarks or an Agent with checkpoint recovery—rather than stopping at a working demo.

The interesting bit

The author treats this as a curriculum, not a blogroll. Projects come with pass/fail criteria that include P99 latency reports, bad-case taxonomies, and budget limits, and the interview section includes real ByteDance Agent-engineering transcripts. It is unapologetically aimed at practitioners who need to ship.

Key highlights

• Staged roadmap from API basics to production engineering and interview prep
• 150+ tutorials and real big-tech interview write-ups (ByteDance, etc.)
• Project rubrics that demand metrics, failure recovery, and cost analysis, not just running code
• Covers recent tooling: MCP, Claude Code, vLLM, SGLang, PagedAttention
• Explicitly excludes large-scale pre-training and fundamental model research

Caveats

• Entirely in Chinese; no English translation is provided
• It is documentation and curated reading lists, not a runnable framework or library
• Not suited for researchers focused on novel architectures or mathematical proofs; the author explicitly warns against using it as the sole source for that path

Verdict

Working programmers with Java, C++, Go, or Python backgrounds who need to bootstrap into LLM application engineering will find this a rare coherent syllabus. If you are looking for source code to import or novel model architectures, look elsewhere.