# LLM-Master

A Chinese-language full-stack learning roadmap for programmers moving into large-language-model application development, covering prompting, RAG, agents, fine-tuning, deployment, Transformer internals and interview preparation.

LLM-Master is a structured curriculum, not a collection of scattered articles, that takes a programmer with a Java, C++, Go, Python or front-end background from basic concepts through application development, retrieval-augmented generation, agents, fine-tuning and deployment, Transformer internals, and interview preparation. Each stage names a learning entry point and a project milestone, such as building a streaming AI application with JSON schema output and cost tracking, or a tool-using agent that can be interrupted, resumed and evaluated.

Topic areas include prompt engineering and structured output; RAG such as chunking, embeddings, vector databases, hybrid retrieval, reranking and evaluation; single- and multi-agent systems such as ReAct, planning, memory and failure recovery; fine-tuning methods such as SFT, RLHF, DPO, LoRA and QLoRA; inference serving with vLLM and SGLang, KV cache and quantization; and a from-scratch Transformer implementation.

The material is regularly updated to track agent harnesses, context engineering, MCP and multi-agent practice.

Website: https://notes.kamacoder.com/llm/