Chapter 43: Vector Databases & RAG with LangChain

You will build a RAG skill BEFORE you learn RAG concepts.

This chapter follows the Skill-First Learning Pattern: In Lesson 0, you create a rag-deployment skill using tools from Chapter 3. Then you spend the chapter understanding what you built and making it better. By the end, you don't just "know RAG"—you OWN a production-ready skill that extends your Task API with semantic search.

The Architecture

This chapter uses a clean separation of concerns:

Layer	Technology	Purpose
Ingestion	LangChain	Document loading, text splitting, embeddings
Storage/Retrieval	LangChain + Qdrant	Vector store, similarity search
Intelligence	OpenAI Agents SDK	Agent orchestration, LLM calls, conversation memory
Evaluation	RAGAS + LangSmith	Quality metrics, tracing

Why this split? LangChain excels at document processing and retrieval. The Agents SDK (Chapter 34) excels at agent orchestration. Together, they give you the best of both worlds without framework lock-in.

The Skill-First Arc

L00: Build Your RAG Skill (20 min)
  |   └── Clone skills-lab, fetch LangChain docs, create skill
  |
  ├── L01-02: Conceptual Foundation
  |   └── Why RAG? + Embeddings mental model
  |
  ├── L03-04: LangChain for Data
  |   └── Document processing, Qdrant vector store
  |
  ├── L05: RAG with Agents SDK
  |   └── Agent + retrieval tool pattern
  |
  ├── L06: Task API Integration
  |   └── Your skill learns semantic task search
  |
  ├── L07: RAG Evaluation
  |   └── LangSmith + RAGAS metrics
  |
  └── L08: Architecture Patterns (Capstone)
      └── 8 RAG patterns—Simple to Agentic

Chapter Structure

#	Lesson	Duration	Your Skill Improves
0	Build Your RAG Skill	20 min	Created from LangChain docs
1	Why Agents Need External Knowledge	25 min	Conceptual: parametric vs retrieval
2	Vector Embeddings Mental Model	30 min	Conceptual: semantic similarity
3	LangChain Document Processing	40 min	Add: chunking strategies
4	Qdrant Vector Store	45 min	Add: vector storage patterns
5	RAG with OpenAI Agents SDK	45 min	Add: agent + retrieval tool pattern
6	RAG for Task API	50 min	Refine: semantic search endpoint
7	Evaluating RAG Quality	45 min	Add: evaluation metrics
8	RAG Architecture Patterns	60 min	Capstone: 2 patterns implemented

Every Lesson Has Skill Reflection

Each lesson ends with "Reflect on Your Skill":

1. Test: Does your skill handle this lesson's concepts?
2. Identify gaps: What's missing?
3. Improve: Add or refine the capability

By Lesson 8, you've tested and improved your skill 8+ times.

Prerequisites

• Chapter 40: FastAPI for Agents (Task API foundation)
• Chapters 34-36: Agent SDK chapters (agents know how to call tools)
• Chapter 3: skill-creator and fetching-library-docs
• Docker: For running Qdrant locally

The Running Example

Your Task API gets smarter. In Chapter 40, you built a Task API with CRUD operations. Now you extend it:

• Lessons 1-5: Learn how RAG works—embeddings, Qdrant, retrieval chains
• Lesson 6: Add /tasks/search/semantic endpoint to your Task API
• Lesson 8: Implement advanced RAG patterns for intelligent task recommendations

The key insight: Semantic search lets users find tasks by meaning, not just keywords.

What You'll Own

.claude/skills/
├── skill-creator/           # From Chapter 3
├── fetching-library-docs/   # From Chapter 3
├── fastapi-agent-api/       # From Chapter 40
└── rag-deployment/          # NEW - this chapter

This skill joins your growing Digital FTE toolkit.

The Mindset Shift

Traditional: "Teach me RAG" Agent Factory: "I own a RAG skill—help me make it better"

You're not here to learn. You're here to build assets.

Start with Lesson 0: Build Your RAG Skill.