limouren2000/llms-dev-study

In plain words:

This repository is a personally curated study guide for people who want to land a job building applications on top of large language models (LLMs) — the tech behind things like ChatGPT. The author's stated goal is the fastest route to employment: the README says the path is designed for “rapid job-landing” and treats efficiency and earning potential as the top priority.

What's inside:

The repo is split into five sections:

1. 0.LLM-Dev Study Router — an overview that explains what an “LLM application developer” role actually is, how it differs from a normal backend developer or an LLM algorithm researcher, and what order a newcomer should learn things in. The core takeaway: the job equals backend development skills plus AI application skills, and this repo focuses on the AI part.

2. 1.RAG — materials about RAG (Retrieval-Augmented Generation), a technique that makes an LLM answer questions by first pulling relevant text from a knowledge base, rather than just relying on what it already knows. This section contains four parts:

   • Two beginner “crash course” videos from Bilibili (a Chinese video site), with the author's own updated code that runs on current package versions. The original authors' code had outdated packages, so the author re-ran and fixed it (marked as the “note” folder). Recommended for quick familiarity.
   • An official LangChain RAG tutorial (llms-3): the author flags this as the most important part because interviews ask about its optimization points. Includes both the original English video and a Chinese translation.
   • An official LangChain sample RAG project (llms-4) that is simple enough to run immediately; the author's modified version lives in a separate repo (chat-langchain-study).
   • Many parts can run on Kaggle, a free cloud-computing platform, and the author links a guide to using Kaggle's free compute.

3. 2.Agent — materials about agents: programs that use an LLM to plan and execute steps (like using tools or searching) instead of just answering queries. This also has four parts:

   • Two simple demo agents from Bilibili tutorials (AI_Agent and QW_Agent). The QW_Agent one requires applying for Qwen (an LLM service) API keys — the author notes there are “pits” (difficulties) and provides pre-fixed code.
   • A 5-day intensive course from Google + Kaggle (dated Nov 10–14, 2025) covering agent basics, tools and the MCP interoperability standard, context engineering, agent quality, and going from prototype to production. Each day has code, official video recordings, and a whitepaper.
   • An entry-level agent project built quickly with MCP (a protocol that lets an LLM connect to external tools/services), in a separate repo called easy-mcp.

4. 3.Interview — interview prep, which the author calls “八股” (a Chinese slang for formulaic knowledge questions). It contains two long notes the author collected while job hunting:

   • A RAG knowledge note, 40,000+ Chinese characters, organized by processing flow.
   • An Agent knowledge note, 10,000+ Chinese characters, organized by module.
   • The author claims these personally helped land “senior AI R&D engineer” roles at major companies..

5. 4.Paper-read — a helper for reading research papers. It provides a Codex skill (a reusable instruction set for OpenAI's Codex tool) called paper-read-brief that condenses ML/DL/multimodal/generative-AI papers into concise Chinese reading notes — extracting core ideas, methods, key figures/tables/formulas, and experiment conclusions section by section. Example notes are included for papers like ReAct, Mem0, Graph of Thoughts, and Tree of Thoughts.

Author's warning about outdated code:

The README explicitly warns that LangChain's official package-version management is chaotic, so some dependencies are probably already outdated. Their suggested fix is pragmatic: paste the error message and relevant code into an AI coding assistant (they name Codex, Claude Code, or Cursor) and let it adapt the code to the current dependency versions.

An online version of the whole guide is available as a web page at limouren2000.github.io/llms-dev-study/ for easier reading.

Bottom line:

This is a job-oriented, heavily curated collection of learning materials — videos, runnable code, interview notes, and paper-reading aids — for anyone aiming to become an LLM application developer, with a strong emphasis on moving fast and avoiding tangents.