gyc-chenxi/llm-fullstack-dev-roadmap

This repository is a structured 100-day study plan for developers who want to build AI applications using large language models. The author wrote it in Chinese, targeting people with a Python background who want to move beyond simply calling AI APIs and instead understand how these systems work at a deeper level. The stated goal is to prepare for job interviews at companies that build or use large language models. The roadmap is split into seven phases. The first few phases cover Python refreshers, prompt design, and how to connect to AI services from providers like OpenAI, DeepSeek, Qwen, and others. It also covers how to track costs, route requests across multiple providers when one is unavailable, and manage access keys securely. By the end of Phase 1, a learner is expected to have built a small chat service that handles multiple providers. Phases 2 and 3 go deeper. Phase 2 walks through how attention mechanisms and transformer architectures actually work, including the math behind them, and covers practical fine-tuning of open models using tools called LLaMA-Factory and vLLM. Phase 3 covers RAG, which is a technique for giving an AI model access to your own documents so it can answer questions about them. It includes handling messy real-world documents like scanned PDFs and tables. Phase 4 is a sprint through 11 well-known open source AI projects, including tools for image generation, multimodal models, and code-writing agents. Phases 5 and 6 cover building AI agents that can take actions and make decisions, and then assembling a production-grade API gateway that routes traffic, enforces rate limits, handles failover, tracks token spending, and can be deployed with Docker. The repository includes Jupyter notebooks and reference tables throughout. It is written almost entirely in Chinese, so readers who do not read Chinese will have limited access to the explanatory content, though the code itself may still be useful.