Langchain

LangChain is a framework for building AI-powered applications using Large Language Models (LLMs).

Without LangChain, you'd need to:

• Write different code for each LLM provider (OpenAI, Anthropic, Azure, etc.)
• Build your own prompt management system
• Create custom tools and function calling logic
• Implement memory and conversation handling from scratch
• Build agent systems without any structure

The LangChain Solution

With LangChain, you get:

• Provider abstraction - Switch between OpenAI, Azure, Anthropic with minimal code changes
• Prompt templates - Reusable, testable prompts
• Tools - Extend AI with custom functions and APIs
• Memory - Built-in conversation history
• Agents - Decision-making AI that can use tools

graph LR
    A[LangChain] --> B[Models]
    A --> C[Prompt Templates]
    A --> D[Tools]
    A --> E[Agents]
    A --> F[Memory]
    B --> G[Provider Abstraction]
    G --> H[OpenAI Models]
    G --> I[Anthropic Models]
    G --> J[Groq Models]
    G --> K[AWS Bedrock Models]
    G --> L[Openrouter Models]
    G --> M[Microsoft Foundry]

Loading

These concepts work together to create powerful AI applications.

---

Environment Setup

Prerequisites

• API Key for any LLM provider e.g. Groq, OpenAI or Openrouter
• Environment variables:
  • GROQ_API_KEY (optional, used for groq provider)
  • OPENAI_API_KEY (optional, used for openai provider)
• Python >= 3.8

Setting Environment From Terminal

1. Install uv with pip install uv
2. Clone repository git clone https://github.com/pasrichashivam/langchain-learning.git
3. Navigate to langchain-learning cd langchain-learning
4. Replicate virtual environment used with working examples uv sync
5. Rename file .env.example to .env and paste your API keys.

---

Core Concepts Overview

LangChain is built around below core concepts:

• Models: AI "brains" that process inputs and generate outputs.
• Prompts: How you communicate with AI models using reusable templates.
• Tools: Extend AI capabilities with external functions and APIs.
• Memory: Remember context across interactions.

How These Concepts Work Together

flowchart LR
    A[User Input] --> B[Memory]
    B --> C[Prompts]
    C --> D[Tools]
    D --> E[Models]
    E --> F[Response]

Loading

---

Messages & Prompts

graph TD
    A[Prompts & Messages] --> B[Messages]
    A --> C[Templates]
    A --> D[Structured Output]
    
    B --> B1[SystemMessage]
    B --> B2[HumanMessage]
    B --> B3[AIMessage]
    B --> B4[ToolMessage]
    
    C --> C1[ChatPromptTemplate]
    C --> C2[PromptTemplate]
    C --> C3[FewShotChatMessagePromptTemplate]
    C --> C4[Template Composition]
    
    D --> D1[Pydantic BaseModel]
    D --> D2[Field Descriptions]
    D --> D3[with_structured_output]
    D --> D4[Nested Schemas]
    
    B4 --> E[Agents]
    C4 --> F[RAG Systems]

Loading

---

Langchain Summary [Still Learning]

Component	Purpose	When to Use
Models	Unified interface for LLMs	Every LLM application
Messages	Manage conversation history	Chat applications
Embeddings	Convert text into vectors	RAG, semantic search
Tools	Connect LLM with external systems	Need APIs or databases
Agents	Autonomous reasoning and tool selection	Multi-step problem solving
Middleware	Intercept requests/responses	Production deployments
Short-Term Memory	Preserve conversational context	Multi-turn chat
Structured Output	Produce validated JSON or typed data	Automation and integrations
Context Engineering	Build the right context for the model	RAG and agent systems
Runtime	Manage execution lifecycle	Production workloads
Streaming	Return tokens incrementally	Interactive user interfaces
MCP	Standard protocol for external tools	Multi-system AI integration
Integrations	Connect to models, vector stores, loaders, and tools	Virtually every application