🌍 For students worldwide · remote trial & live lessons available (time zones arranged)
Home/ Courses/ RAG Retrieval Augmentation
LLM ApplicationsAdvancedUpdated 2026-06-01

RAG Retrieval Augmentation

32 lessons 16 hours 32 videos / articles

RAG (retrieval-augmented generation) helps large language models answer more accurately and usefully by retrieving relevant information from an external knowledge base. This course is built around real-world RAG systems: you first understand how retrieval and generation cooperate, then learn the key trade-offs in vector databases, hybrid search, prompt design, system evaluation, and production deployment.

By the end of this course you will be able to:

  • Understand how retrieval and generation cooperate, and design reliable, evaluable RAG systems
  • Master common retrieval techniques — vector search, hybrid search, BM25 + ANN
  • Judge which embedding model / vector database fits each scenario and select correctly
  • Make steadier production engineering decisions around latency, cost, and context windows
  • Complete an enterprise knowledge-base end-to-end project you can present in interviews
Course consult
Enroll via Xiaohongshu

📕 DM 「兔老板工作室」 on Xiaohongshu to enroll now

Enroll by DM · no platform payment · always valid

The course is organized into 5 modules — 32 lessons in total.

01 RAG Foundations & Architecture
6 lessons · 3 hours
+
  • Why RAG: the three limits of LLMs28 min
  • RAG architecture overview: retriever / knowledge base / LLM / evaluation30 min
  • Vector retrieval internals: ANN algorithms HNSW / IVF32 min
  • Hybrid search: BM25 + vector retrieval28 min
  • RAG paradigm evolution: Naive / Advanced / Modular30 min
  • Foundations module quiz: 12 high-frequency interview questions28 min
02 Vector Databases & Embeddings
8 lessons · 4 hours
+
  • Milvus architecture & index selection32 min
  • Qdrant / Weaviate / pgvector compared30 min
  • Choosing embedding models: BGE / M3E / OpenAI32 min
  • Embedding fine-tuning for domain adaptation34 min
  • Chunking strategies: fixed / semantic / hierarchical30 min
  • Document preprocessing: cleaning, structuring, tables28 min
  • Vector DB module quiz: 15 high-frequency interview questions32 min
03 Advanced Retrieval Optimization
6 lessons · 3 hours
+
  • Reranking: BGE-reranker / Cohere30 min
  • Query rewriting: HyDE / Multi-Query / Step-Back32 min
  • Multi-hop retrieval & sub-question decomposition28 min
  • Context compression & information-density optimization26 min
  • Retrieval optimization quiz: 10 high-frequency interview questions28 min
04 Generation & End-to-End Project
6 lessons · 3 hours
+
  • Prompt templates: citation grounding & format control26 min
  • Streaming generation & frontend integration28 min
  • Enterprise knowledge-base end-to-end project (part 1)38 min
  • Enterprise knowledge-base end-to-end project (part 2)40 min
  • Performance optimization & cost reduction28 min
  • Project module quiz: 10 high-frequency interview questions26 min
05 Evaluation System
6 lessons · 3 hours
+
  • RAGAS evaluation metrics explained30 min
  • TruLens / DeepEval compared28 min
  • How to build a domain evaluation set32 min
  • A/B testing & production monitoring28 min
  • Closing the evaluation loop & continuous optimization26 min
  • Evaluation module quiz: 8 high-frequency interview questions24 min

🎯 Who it’s for

  • Engineers starting out or leveling up in LLM application development
  • Job seekers interviewing for LLM application / RAG roles
  • Backend engineers pivoting toward LLM applications
  • Engineers already doing RAG who want to systematize it

🚀 What you’ll gain

  • The ability to design complete industrial-grade RAG systems
  • Confident selection of mainstream vector DBs / embedding models
  • One presentable, end-to-end enterprise-grade RAG project
  • Model answers to 55 high-frequency RAG interview questions
  • An engineering mindset for evaluation loops & continuous optimization

📋 Prerequisites

  • Comfortable with Python async programming
  • Basic familiarity with LLM API calls (OpenAI / Qwen)
  • No vector DB experience needed — start from zero

⏰ Pace

  • 16 hours total; plan 3–4 weeks
  • 8 lessons a week + hands-on project time on weekends
  • Access starts immediately on enrollment; review anytime, forever
兔老板

The Instructor

CAS PhD · senior algorithm engineer · sits on real hiring loops · author behind the WeChat account 「兔老板工作室」

View the full instructor profile →

Engineers who want to start or level up in LLM application development, candidates interviewing for LLM application / RAG roles, and backend engineers pivoting into LLM applications. Prerequisites are simply comfort with Python async programming and basic familiarity with LLM API calls (OpenAI / Qwen). No vector DB experience is required — you start from zero.
Five modules across 32 lessons: RAG foundations & architecture, vector databases & embeddings, retrieval optimization, generation & an end-to-end project, and an evaluation system. It covers 5 optimization strategies (query rewriting / Rerank / HyDE / context compression / multi-hop), industrial-grade vector DB selection (Milvus / Qdrant / pgvector), embedding fine-tuning, and an evaluation loop with RAGAS / TruLens / DeepEval. You complete an enterprise knowledge-base end-to-end project, plus detailed solutions to 55 high-frequency RAG interview questions.
32 lessons, about 16 hours, in 5 modules. We recommend finishing in 3–4 weeks (8 lessons a week + weekend project practice). Access opens immediately after enrollment, and you can review it forever.
RAG doesn’t change model weights: it retrieves fresh or private information from an attached knowledge base and appends it to the context — ideal when knowledge updates quickly or the data is private. Fine-tuning modifies the weights, better for internalizing style, format, or domain capability. The two aren’t in conflict and are often combined in industry; interviewers frequently probe engineering judgment by asking “why RAG instead of fine-tuning?”
High-frequency topics: query rewriting, Rerank, HyDE, vector DB selection (Milvus / Qdrant / pgvector), embedding & chunking strategy, context compression, multi-hop retrieval, and evaluation loops such as RAGAS. Module 5 pairs the course with 55 high-frequency RAG interview questions in depth, from principles to project follow-ups.
Module 4 walks you through an enterprise knowledge-base end-to-end project — data cleaning, embeddings, retrieval optimization, and the evaluation loop. Even without production experience, you’ll be able to explain the reasoning behind each choice and trade-off. Interviewers value your “why did you do it this way” decision-making more than whether a RAG project appears on your résumé.

RAG is a must-know for LLM application roles

The go-to intro course — follow 「兔老板工作室」 on Xiaohongshu and consult to enroll now.

Enroll via Xiaohongshu
🎓 More systematic courses
Trial first, then enroll · DM 「兔老板工作室」 on Xiaohongshu to consult
📘 Full-Stack LLM Interview Breakthrough Program 🤖 AI Agent Bootcamp ⚙️ LLM Fine-tuning & Alignment ⚡ Transformer & Inference Optimization 🎯 Resume + mock interview 1-on-1
📚 Free guides — read them right here: LLM algorithm-role high-frequency checklist · High-frequency RAG interview Q&A · Agent system design · Big-tech talent programs · LLM fine-tuning & alignment points · LLM inference optimization points · Résumé & project pitfall guide · AI Infra free practice question bank