LangChain

Status

✅ Notes complete — all subtopics filled in from the CampusX LangChain course (~511 pp), official docs, and the CampusX GitHub repos.

Reading order

If you're new, read in order — each topic builds on the previous.

#	Topic	Read
1	Introduction — what LangChain is and why	01-introduction
2	Models — Chat + Embedding	02-models
3	Prompts & Templates	03-prompts
4	Output Parsers & Structured Output	04-output-parsers
5	Chains — Sequential, Parallel, Branching	05-chains
6	Runnables & LCEL — the engine underneath	06-runnables-and-lcel
7	Document Loaders	07-document-loaders
8	Text Splitters	08-text-splitters
9	Vector Stores & Embeddings	09-vector-stores
10	Retrievers	10-retrievers
11	RAG — end-to-end	11-rag
12	Memory & Conversation State	12-memory
13	Tools & Tool-Calling	13-tools
14	Agents (ReAct loop)	14-agents

The big picture

Three layers, learned in three parts:

flowchart TB
    subgraph SG1 [Part 1 — Fundamentals 1-6]
      A1[Models] --> A2[Prompts]
      A2 --> A3[Output Parsers]
      A3 --> A4[Chains + LCEL]
    end
    subgraph SG2 [Part 2 — RAG 7-11]
      B1[Document Loaders] --> B2[Text Splitters]
      B2 --> B3[Vector Stores]
      B3 --> B4[Retrievers]
      B4 --> B5[RAG end-to-end]
    end
    subgraph SG3 [Part 3 — Agentic 12-14]
      C1[Memory] --> C2[Tools] --> C3[Agents]
    end
    A4 -.is the spine of.-> B5
    A4 -.is the spine of.-> C3

Learning roadmap

• Core primitives — prompts, chat models, output parsers
• LCEL (| pipe operator) and Runnable interface
• Memory & message history
• Document loaders & text splitters
• Embeddings & vector stores
• Retrievers & RAG (retrieval-augmented generation)
• Tools & tool-calling
• Agents (create_react_agent)
• Streaming & basic production patterns
• Integrations: OpenAI, Anthropic, Ollama, Pinecone, Chroma

What I should be able to do after this topic

• Compose any LLM pipeline using LCEL without copy-pasting boilerplate.
• Build a working RAG system from scratch (load → split → embed → store → retrieve → answer with citations).
• Choose between LangChain, raw API calls, and LangGraph based on the task.
• Build a small ReAct agent with custom tools.
• Stream responses and handle errors gracefully.