Response time slows down
Blindly attaching search, reranking, summarization, and verification steps may improve accuracy, but it will cause service latency to increase drastically.
🔎 Classic RAG
🕸️ Graph RAG
🤖 Agentic RAG
🧠 xMemory
Going beyond simple search,
we design a self-reasoning RAG system.
This is a practical course where you will clearly understand the differences between Classic RAG, Graph RAG, and Agentic RAG. You will implement a dynamic routing system that selects the optimal pipeline based on query characteristics, as well as a long-term memory architecture based on GraphRAG and xMemory.
28 Lectures
Approx. 2 hours 34 minutes
3 types
of RAG architecture comparison
Hybrid
Dynamic Routing Design
xMemory
Long-term Memory System
🚨 Are you still designing RAG in only one way?
Accuracy, cost, and latency
always fluctuate together.
RAG systems are not just about putting documents into a vector DB and searching them. In an actual service, completely different designs are required depending on the nature of the question, document structure, cost, response speed, and depth of reasoning.
Token costs will increase
Simply including a lot of context for every question is straightforward, but it is difficult to avoid operational costs and context waste.
It is weak against complex questions
Simple similarity search alone makes it difficult to sufficiently handle relationships between multiple documents, cause and effect, and long-term context.
좋은 RAG는 “더 많이 검색하는 시스템”이 아닙니다.
It is a system that determines the nature of a question and selects the most appropriate search, reasoning, and verification strategies. This course covers both the criteria for that judgment and the implementation structure.
🧠 What you will gain after taking this course
The power to personally judge and design
RAG architectures
Beyond simply typing along with basic implementation examples, this course focuses on developing the design intuition to independently determine which RAG structure to choose for any given situation.
Architecture Selection Criteria
Classic, Graph, and Agentic RAG's strengths, weaknesses, and suitable use cases can be clearly distinguished.
Implementing Dynamic Routing
Design a hybrid routing structure that selects fast retrieval, deep reasoning, or self-evaluation loops based on the question type.
Building a Long-Term Memory System
Learn advanced RAG systems featuring relational network reasoning and continuous memory structures by combining GraphRAG and xMemory.
🧱 Core Architecture Comparison
From Classic to Agentic,
summarizing the evolution of RAG all at once.
Classic RAG
This is the basic structure for quickly generating answers based on vector search and context injection. It is effective for simple questions, FAQs, and document-based responses.
- Document Chunking and Embedding
- Hybrid Search
- Token Diet
Graph RAG
It extracts entities and relationships from the text to create a knowledge graph and performs deeper reasoning by utilizing the connections between documents.
- Knowledge Graph Extraction
- Network-based Search
- Handling Complex Questions
Agentic RAG
AI evaluates the search results, searches again if they are insufficient, and performs self-checks on the quality of the answer in an autonomous search loop.
- Self-evaluation loop
- Iterative Search Strategy
- Answer Verification Flow
🚀 Practical Design Flow
Look at the question and
select the optimal RAG pipeline.
Not all questions are processed in the same way. We design a structure that balances cost and quality by distinguishing between questions that require a quick answer, those that require relational reasoning, and those that require self-evaluation.
STEP 1
Query Analysis
Determine the complexity of the question, the required context, and the depth of reasoning.
STEP 2
Router Selection
Dynamically selects the most appropriate path among Classic, Graph, and Agentic.
STEP 3
Search & Reasoning
It executes document retrieval, relationship network exploration, and self-evaluation loops according to the situation.
STEP 4
Cost Optimization
Reduces token usage and latency while maintaining answer quality.
✅ Recommended for the following people
Those who want to elevate RAG from the “implementation”
to the “design” stage
AI Engineer
It is suitable for those who want to solve accuracy, token cost, and latency issues in existing RAG systems and design better retrieval and reasoning structures.
Backend Developer
It is ideal for those who want to enhance their architecture design capabilities for complex knowledge management systems, document retrieval systems, and LLM-based services.
AI Product Manager
Recommended for those who want to understand the differences in RAG technologies and determine which architecture is necessary for production-level LLM products.
📚 Curriculum at a Glance
From the evolution of RAG
to production-ready hybrid architectures
Section 1
Evolution of the RAG Paradigm and Overview of Architecture Design
Summarize the RAG trilemma, structural differences between Classic/Graph/Agentic RAG, and the necessity of hybrid routers.
Section 2
Advanced Classic RAG and Dynamic Routing Systems
Expand basic RAG into a practical version through Advanced Classic RAG, token dieting, and GraphRAG adoption criteria.
Section 3
Graph RAG: Relational Network Reasoning using Knowledge Graphs
Extract knowledge graphs from text and learn the structure of a Hybrid Searcher that utilizes relational networks.
Section 4
Agentic RAG: Self-Evaluation and Autonomous Search Loops
It covers the core loop of Agentic RAG, which evaluates search results on its own and re-searches for missing information.
Section 5
Building Advanced RAG Based on Long-Term Memory Systems (xMemory)
Build an advanced RAG structure capable of long-term memory and relational reasoning by combining GraphRAG and xMemory.
Section 6+
Hybrid RAG Architecture for Production Environments
It expands the perspective on how to organize knowledge in a practical environment and how to design the paths for retrieval and reasoning.
🔎 Notes before taking the course
If you know this much,
you will be able to absorb the content much faster.
- Basic knowledge of Python programming is required.
- It is helpful to have an understanding of prompt engineering and basic RAG concepts.
- If you have experience integrating LLM APIs based on Python, it will be easy to follow the flow of the hands-on exercises.
- It will be even more beneficial if you have experience using vector databases or search systems.
RAG is no longer just a simple search function,
but the core architecture of AI products.
Quickly search with Classic RAG, understand relationships with Graph RAG, evaluate autonomously with Agentic RAG, and design long-term memory with xMemory. This course will give you the insight to view RAG systems as practical architectures.
Classic RAG
Graph RAG
Agentic RAG
Hybrid Router
Token Diet
xMemory
Long-term Memory
It is great to be able to understand the paradigm shift of RAG and the concept of hybrid systems, while also learning various ways to optimize RAG.
This is an excellent lecture that provides a clear understanding of the new trends and overall architecture of RAG. At first, I thought I understood everything in my head, but I felt a bit lost and empty after each section, wondering, "Wait, how do I actually do this?" However, seeing how the instructor immediately incorporated the feedback and improvement requests I provided, I felt once again that this is a great instructor and a high-quality course. I am looking forward to the rest of the course even more as practice code and additional content are being added to each section. The content is fantastic, the explanations are clear, and the instructor is very responsive to feedback, so I highly recommend taking this course!
Overall, I think it was a great lecture. In particular, I have already developed a RAG service that retrieves answers using BM25+Vector, and I feel I can improve it further based on this course. However, one thing I wish there was more of is content on embedding techniques. I am currently chunking text or Markdown documents and storing them in a Qdrant vector database; I was hoping for some tips and know-how on how to efficiently store chunks and vector data, as well as how to retrieve that data effectively. Still, it was great to learn advanced technologies like GraphRAG and Agentic RAG!
![Just 1 hour! Creating 'My Own AI Senior Developer' to install on my computer (Antigravity Vibe Coding) [Source code provided]Course Thumbnail](https://cdn.inflearn.com/public/files/courses/340332/cover/ai/3/e87ee52b-1099-42db-a384-64ab8c725470.png?w=420)
![[Beginner] Mastering Qt Usage Through 6 Projects (REST API)Course Thumbnail](https://cdn.inflearn.com/public/courses/333527/cover/eb60879b-b2d9-4194-bbb0-9291102ceda3/333527.png?w=420)
![[Inflearn Award Bestseller] How to Become an AI Automation Expert Without Coding, The Complete n8n GuideCourse Thumbnail](https://cdn.inflearn.com/public/files/courses/337053/cover/01jrkjt8emye2z7j7846br3vkk?w=420)
