🤖 Agentic Notebooks (LangGraph + LangChain)

Hands-on Jupyter notebooks covering agents, tool-calling, and RAG (Retrieval-Augmented Generation) using LangGraph/LangChain with multiple vector store backends.

📌 What You’ll Find Here

• Build blocks: prompts, chains, embeddings, loaders, and vector stores
• RAG: chunking → embedding → retrieval → generation
• Agent patterns: tool calling, routing, memory, and “corrective” retrieval loops

🗺️ Notebook Guide (Crisp Map)

🧩 LangGraph (agent graphs, tools, state)

Notebook	What it covers	Key concepts
langgraph_intro.ipynb	First steps with StateGraph and state passing	Typed state, node functions, message/state updates
langgraph_Agentic_Class_2.ipynb	Embeddings + Chroma retrieval wired into an agent-style state	HuggingFace embeddings, Chroma retriever, structured state
langgrapha_Agentic_Class_3.ipynb	Graph with and without tool calling	ToolNode, custom tools, message history, memory patterns
langgraph_Agentic_Class_4.ipynb	Tool binding and tool selection behavior	bind_tools, inbuilt search tools, tool routing conditions
tools.ipynb	Minimal custom tool examples	@tool, tool schemas, tool invocation
Agentic RAG Class 5.ipynb	End-to-end RAG from web content	Web loaders, splitting, embedding, retrieval, generation
Corrective RAG Class 5.ipynb	“Corrective RAG”: grade → rewrite → fallback search	Relevance grading, JSON parsing, query rewrite, web search fallback
Network Multi Agent_Class_6.ipynb	Advanced agent behaviors with tools (and demos)	ReAct-style agent usage, commands, tool execution, search/code tools
Supervisor Multi Agent Class 7.ipynb	Supervisor agent orchestrating specialist agents	supervisor agent, delegation, role-based agents
Human-In-Loop Class 8.ipynb	Human-in-the-loop interventions in agent workflows	human feedback, approvals, intervention points

🧱 LangChain (foundations, ingestion, vector DBs)

Notebook	What it covers	Key concepts
basics.ipynb	Core LangChain building blocks	prompts, chains, model invocation, LangSmith env vars
embeddings.ipynb	Embedding text into vectors	HuggingFaceEmbeddings, vector dimensions, query vs doc embeddings
data-ingestion.ipynb	Loading documents from files	TextLoader, PDF loaders, lazy loading, document objects
vectoredatabase.ipynb	Vector similarity search + RAG with FAISS	cosine vs L2, FAISS indexing, retrieval, RAG patterns
code.ipynb	Pinecone-backed vector store	index creation, upsert, query, PineconeVectorStore

✅ Pydantic (structured outputs & validation)

Notebook	What it covers	Key concepts
pydantic.ipynb	Data validation vs plain dataclasses	BaseModel, type coercion, ValidationError, optional fields

🤝 AutoGen (multi-agent workflows and tools)

Notebook	What it covers	Key concepts
first autogen agent.ipynb	Your first AutoGen agent with a basic task	single agent, prompts, run loop
Autogen Agents.ipynb	Core AutoGen agent patterns	user proxy, assistant agents, conversations
Autogen Agents Tools.ipynb	Using tools with AutoGen agents	tool calling, code execution, retrieval tools
Autogen Teams.ipynb	Teams of cooperating agents	multi-agent teams, roles, coordination
Autogen Graph Flow.ipynb	Graph-style workflows in AutoGen	workflow graphs, routing, multi-step tasks
Selection Group Chat.ipynb	Group chat with agent selection	group chat, speaker selection, collaboration
Human in the loop.ipynb	Human supervision in AutoGen flows	human approvals, intervention, oversight
Human in the loop Outside.ipynb	Human-in-the-loop with external context	external tools, human feedback, review loops
mcp_in_autogen.py	Connecting AutoGen agents to MCP servers	McpWorkbench, StdioServerParams, tool discovery

🛠️ Agent Development Kit (ADK)

File	What it covers	Key concepts
built_in_tool_agent/agent.py	Multi-agent orchestration with built-in tools	AgentTool, google_search, agent wrapping
crew_ai_tool_agent/agent.py	Integrating CrewAI tools into ADK	CrewaiTool, SerperDevTool, tool wrapping
custom_tool_agent/agent.py	Custom Python functions as agent tools	Function tools, LlmAgent, tool definitions
openrouter_agent/agent.py	Using external LLMs via OpenRouter	LiteLlm, OpenRouter integration, multi-provider support
session_runner/runner.py	Persistent session management and state	Runner, InMemorySessionService, stateful agents
structred_output/agent.py	Structured data extraction with Pydantic	output_schema, BaseModel, data validation

🏗️ Architecture Diagrams

🤖 AutoGen Team Patterns

AutoGen supports various ways to coordinate multiple agents:

1. Round Robin (Fixed Sequence)

Agents speak in a pre-defined circular order.

graph LR
    User([User]) --> Team[RoundRobinGroupChat]
    Team --> A[Agent 1]
    A --> B[Agent 2]
    B --> C[Agent 3]
    C --> A

Loading

2. Selector (Dynamic Selection)

A "Selector" LLM determines the next best agent to speak based on their description.

graph TD
    User([User]) --> Team[SelectorGroupChat]
    Team --> Selector{Selector LLM}
    Selector -->|analyzes task| A[Planning Agent]
    Selector -->|analyzes task| B[Worker Agent]
    Selector -->|analyzes task| C[Analyst Agent]
    A & B & C --> Team

Loading

3. Graph Flow (State Machine)

Sequence is defined by a Directed Graph (DiGraph) with explicit edges.

graph LR
    User([User]) --> Graph[DiGraphBuilder / GraphFlow]
    Graph --> Writer[Writer Agent]
    Writer --> Reviewer[Reviewer Agent]
    Reviewer -- feedback --> Writer
    Reviewer -- approved --> End([End Task])

Loading

🛠️ ADK Core Architecture

The Google Agent Development Kit (ADK) focuses on session management and persistent runners.

graph TD
    User([User]) --> Runner[Runner]
    subgraph ADK Framework
        Runner --> Session[InMemorySessionService]
        Runner --> Agent[Agent / LlmAgent]
        Session <-->|State/Context| Agent
        Agent --> Tools[Tools Library]
        subgraph Tools Library
            Tools --> BuiltIn[Google Search / etc.]
            Tools --> Custom[Python Functions]
            Tools --> CrewAI[CrewAI Tool Wrappers]
        end
    end
    Agent --> LLM[Gemini / LiteLLM]

Loading

🔁 Visual: Corrective RAG Loop (High Level)

flowchart LR
  Q[User Question] --> R[Retrieve from Vector Store]
  R --> G[Grade Relevance]
  G -->|relevant| A[Answer with Context]
  G -->|irrelevant| W[Rewrite Query]
  W --> S[Web Search]
  S --> R

Loading