LangChain Agent Project

A comprehensive collection of LangChain and LangGraph examples demonstrating AI agents, tool calling, real-time web search, and RAG (Retrieval Augmented Generation) with vector databases.

🚀 Features

• Text Summarization with OpenAI and local Ollama
• Tool Calling with static functions
• Real-time Web Search with Tavily API
• Structured Responses with Pydantic models
• LangGraph ReAct Agents with multi-step reasoning
• RAG with Vector Databases using ChromaDB
• MongoDB Atlas Vector Search for cloud-based RAG
• Short-Term Memory with conversation chains
• Job Search Agent example

📋 Prerequisites

• Python 3.9+
• uv package manager
• Ollama (for local LLM)
• API Keys:
  • OpenAI API key (optional, for GPT models)
  • Tavily API key (for real-time search)

🔧 Installation

1. Clone the repository

git clone <your-repo-url>
cd LangChainProject

2. Install dependencies

uv sync

3. Set up environment variables

Create a .env file in the project root:

OPENAI_API_KEY=sk-proj-your-key-here
TAVILY_API_KEY=tvly-dev-your-key-here

4. Install and run Ollama (for local models)

# Install Ollama from https://ollama.ai/
# Pull the llama3.1 model
ollama pull llama3.1:8b

📁 Project Structure

LangChainProject/
├── main_1summerization.py              # Text summarization with OpenAI
├── main_2summerization_with_local_ollama.py  # Summarization with Ollama
├── main_3tool_call.py                  # Static tool calling example
├── main_4tool_call_tavily.py           # Multiple tools + Tavily search
├── main_5simple_tavily.py              # Simple Tavily search with structured output
├── main_6job_search_example.py         # Job search agent example
├── main_7langgraph_react_agent.py      # LangGraph ReAct agent (NEW! ⭐)
├── main_8chromadb_rag.py               # Knowledge base RAG (in-memory)
├── main_9real_chromadb.py              # Real ChromaDB vector database (NEW! ⭐)
├── main_10_mongo_vector_search.py      # MongoDB Atlas Vector Search (NEW! ⭐)
├── main_11_memory_example.py           # Short-term memory example (NEW! ⭐)
├── flow_7.png                          # LangGraph visualization
├── .env                                # Environment variables (create this)
├── pyproject.toml                      # Project dependencies
├── chroma_db/                          # ChromaDB storage (created on first run)
│   ├── chroma.sqlite3                  # SQLite metadata
│   └── [UUID]/                         # Vector embeddings
└── README.md                           # This file

🎯 Examples

1. Text Summarization with OpenAI

Summarizes text about Ruby Learner platform using GPT-3.5-turbo.

uv run python main_1summerization.py

Output: A concise 2-bullet summary of the input text.

---

2. Text Summarization with Local Ollama

Same as above but uses local Ollama model (no API key needed).

uv run python main_2summerization_with_local_ollama.py

Advantages:

• 🆓 Free - no API costs
• 🔒 Private - runs locally
• ⚡ Fast - no network latency

---

3. Static Tool Calling

Demonstrates how AI agents can call custom functions to retrieve information.

uv run python main_3tool_call.py

Features:

• Static course database
• get_course_info() function
• get_student_count() function
• LLM decides which tool to use

Example Query: "Tell me about the Flutter course and how many students it has."

---

4. Multiple Tools + Tavily Search

Combines static tools with real-time web search using Tavily.

uv run python main_4tool_call_tavily.py

Features:

• Tests 3 different queries:
  1. Latest AI news (uses Tavily)
  2. Course information (uses static function)
  3. Bitcoin price (uses Tavily)
• LLM automatically chooses the right tool

Requires: TAVILY_API_KEY in .env file

---

5. Simple Tavily Search ⭐ (Recommended)

Clean, simple example of real-time web search with structured responses.

uv run python main_5simple_tavily.py

Features:

• Real-time web search
• Structured output with Pydantic models
• Includes sources (URLs and titles)
• Easy to customize for your queries

Example Output:

Answer:
The latest news about AI in December 2025 includes Google's new Gemini 3 Deep Think mode...

Sources:
  1. AI News Today, December 5, 2025...
     https://ts2.tech/en/ai-news-today-december-5-2025...
  2. Anthropic News Today...
     https://ts2.tech/en/anthropic-news-today...

---

6. Job Search Agent

Searches for job postings on LinkedIn and returns structured results.

uv run python main_6job_search_example.py

Features:

• Searches for AI Engineer jobs
• Returns company, title, location, description
• Includes source URLs
• Similar to the reference example pattern

Example Query: "Search for 3 job postings for an AI engineer using langchain in the bay area on linkedin"

---

7. LangGraph ReAct Agent ⭐ (NEW!)

Multi-step reasoning agent that can use multiple tools and decide when to stop.

uv run python main_7langgraph_react_agent.py

Features:

• 🔄 ReAct Pattern: Reasoning → Action → Observation loop
• 🛠️ Multiple Tools: Calculator + Tavily web search
• 🤖 Agent Decides: When to use tools, which tools, and when to stop
• 📊 Visualization: Generates flow_7.png showing the agent workflow

Key Difference from LangChain:

• LangChain: Single pass, one-shot tool calling
• LangGraph: Multi-step reasoning, can call tools multiple times

Example Queries:

# Query 1: Math calculation
"What is 1234 multiplied by 5678?"

# Query 2: Web search + reasoning  
"What is the temperature in Tokyo? List it and then triple it"

Why LangGraph?

• More control over agent flow
• Can create complex multi-agent systems
• Better for tasks requiring multiple steps
• Similar to how ChatGPT/Claude work internally

---

8. Knowledge Base RAG (In-Memory)

Simple RAG example using in-memory knowledge base (no database setup needed).

uv run python main_8chromadb_rag.py

Features:

• 📚 Stores course information (Flutter, Kotlin, LangChain, AI/ML)
• 🔍 Agent searches knowledge base before answering
• 👥 Includes metadata (student counts, topics)
• 💡 Perfect for learning RAG concepts

Example Queries:

# Query 1
"Tell me about Flutter and how many students are enrolled?"

# Query 2  
"What courses are available about mobile development and AI?"

---

9. Real ChromaDB Vector Database ⭐ (NEW!)

Production-ready RAG with persistent vector database using ChromaDB.

uv run python main_9real_chromadb.py

Features:

• 💾 Persistent Storage: Data saved to disk (./chroma_db/)
• 🔢 Vector Embeddings: Uses Ollama to generate 4096-dimensional vectors
• 🎯 Semantic Search: Finds similar content, not just keyword matching
• ⚡ Fast: HNSW algorithm for O(log n) similarity search
• 🗄️ SQLite Backend: Metadata stored in SQLite, vectors in binary files

What Gets Created:

chroma_db/
├── chroma.sqlite3          # Metadata (258 KB)
└── [UUID]/
    └── data_level0.bin     # Vector embeddings (1.6 MB)

How It Works:

1. First run: Creates ChromaDB, generates embeddings (takes ~30 seconds)
2. Later runs: Loads from disk instantly
3. Agent searches by semantic similarity
4. Returns relevant documents with metadata

Example Output:

🔧 Creating new ChromaDB...
⏳ Generating embeddings with Ollama...
✅ ChromaDB created at: /path/to/chroma_db
📊 Stored 5 documents with vector embeddings

Query: Tell me about mobile app development frameworks

🔍 Searching ChromaDB for: 'mobile app development frameworks'
✅ Found: Flutter, Kotlin, LangChain courses

Why Use ChromaDB?

• ✅ Store your company's knowledge
• ✅ Semantic search (understands meaning, not just keywords)
• ✅ Scales to millions of documents
• ✅ Perfect for chatbots, documentation search, Q&A systems

See chromadb_structure.md for detailed architecture explanation.

---

10. MongoDB Atlas Vector Search ⭐ (NEW!)

Cloud-based vector search using MongoDB Atlas with LangChain integration.

uv run python main_10_mongo_vector_search.py

Features:

• ☁️ Cloud Database: MongoDB Atlas (no local setup needed)
• 🔢 Vector Embeddings: Uses Ollama nomic-embed-text model
• 🌐 Atlas Vector Search: Native MongoDB vector search capabilities
• 🔄 Fallback Search: Manual cosine similarity if Atlas search index not configured
• 📊 Metadata Support: Stores and queries course info, topics, student counts

Prerequisites:

1. MongoDB Atlas account (free tier available)
2. Connection string with password in .env:

   MONGODB_URI=mongodb+srv://user:password@cluster0.mongodb.net/...
   MONGO_DB_NAME=rab_test
   MONGO_COLLECTION=rab_test

What Gets Created:

MongoDB Atlas:
└── rab_test (database)
    └── rab_test (collection)
        ├── Document 1 (page_content, embedding, course, topic, students)
        ├── Document 2 (...)
        └── Document 3 (...)

How It Works:

1. Connects to MongoDB Atlas cloud database
2. Generates embeddings using Ollama locally
3. Stores documents with vector embeddings in MongoDB
4. Performs semantic similarity search
5. Returns relevant documents with scores

Example Output:

📌 Query: Tell me about mobile app development frameworks

   Result 1 (similarity: 0.8120):
   Course: Flutter
   Topic: Mobile Development
   Students: 150
   Content: Flutter is a cross-platform mobile development framework...

   Result 2 (similarity: 0.7243):
   Course: Kotlin
   Topic: Android Development
   Students: 120
   Content: Kotlin is a modern programming language...

ChromaDB vs MongoDB Atlas:

Feature	ChromaDB (main_9)	MongoDB Atlas (main_10)
Storage	Local files	Cloud database
Setup	Zero config	Requires MongoDB account
Scalability	Limited to disk	Unlimited cloud storage
Multi-user	Single machine	Multiple users/apps
Cost	Free	Free tier available
Best For	Development, prototypes	Production, team collaboration

When to Use MongoDB Atlas:

• ✅ Production applications
• ✅ Team collaboration (shared database)
• ✅ Already using MongoDB for other data
• ✅ Need backup/replication
• ✅ Want managed infrastructure

---

11. Short-Term Memory Example ⭐ (NEW!)

Interactive conversation with memory using ConversationBufferWindowMemory.

uv run python main_11_memory_example.py

Features:

• Interactive Chat: Type messages and get AI responses
• Short-Term Memory: Remembers last K interactions (configurable)
• Automatic Updates: Memory updates with each conversation turn

How It Works:

1. Uses ConversationBufferWindowMemory to store recent messages
2. ConversationChain automatically manages memory context
3. Each response considers previous conversation history
4. Memory window prevents unlimited growth (keeps only recent interactions)

Example Interaction:

You: Hi, my name is John
AI: Hello John! Nice to meet you.

You: What's my name?
AI: Your name is John.

Memory Types in LangChain:

• ConversationBufferMemory: Unlimited memory (grows forever)
• ConversationBufferWindowMemory: Limited to last K interactions ⭐
• ConversationSummaryMemory: Summarizes old messages to save space
• ConversationTokenBufferMemory: Limited by token count

Why Short-Term Memory?

• Prevents memory from growing too large
• Focuses on recent context
• Better for long conversations
• Configurable window size

---

🔑 API Keys

Get Tavily API Key

1. Visit https://tavily.com/
2. Sign up for free account
3. Get your API key
4. Add to .env file:

   TAVILY_API_KEY=tvly-dev-your-key-here

Get OpenAI API Key (Optional)

1. Visit https://platform.openai.com/
2. Create account and add payment method
3. Generate API key
4. Add to .env file:

   OPENAI_API_KEY=sk-proj-your-key-here

Note: You can use Ollama (free, local) instead of OpenAI for most examples.

---

📊 Quick Comparison

Feature	LangChain (main_1-6)	LangGraph (main_7)	RAG (main_8-9)	MongoDB Atlas (main_10)	Memory (main_11)
Purpose	Single-pass tasks	Multi-step reasoning	Knowledge retrieval	Cloud vector search	Conversation context
Tool Calls	One-shot	Multiple, iterative	Search database	Search cloud DB	N/A
State	Stateless	Stateful graph	Persistent data	Cloud persistent	Conversation history
Storage	N/A	N/A	Local files	MongoDB Atlas	In-memory buffer
Use Case	Simple queries	Complex workflows	Company knowledge	Production RAG	Chatbots, assistants
Example	"What's 2+2?"	"Search, then calculate"	"Find in our docs"	"Team knowledge base"	"Remember my name"

When to Use What?

• LangChain (main_1-6): Quick tasks, single tool calls, real-time data
• LangGraph (main_7): Multi-step reasoning, agent workflows, complex decisions
• RAG ChromaDB (main_8-9): Local development, prototypes, single-user apps
• RAG MongoDB (main_10): Production apps, team collaboration, cloud deployment
• Memory (main_11): Interactive applications, chatbots, virtual assistants

---

🧠 Learning Path

Beginner:

1. main_1 → Learn basic LangChain
2. main_2 → Try local Ollama
3. main_3 → Understand tool calling

Intermediate: 4. main_5 → Real-time search with Tavily 5. main_7 → Multi-step agents with LangGraph 6. main_8 → Understand RAG concepts

Advanced: 7. main_9 → Production RAG with ChromaDB (local) 8. main_10 → Cloud RAG with MongoDB Atlas (production) 9. main_11 → Interactive applications with memory

---

🛠️ Troubleshooting

Ollama Connection Error

httpcore.ConnectError: [Errno 61] Connection refused

Solution: Start Ollama first

ollama serve

Tavily API Error

ValueError: TAVILY_API_KEY not found

Solution: Add key to .env file

TAVILY_API_KEY=tvly-dev-your-key-here

ChromaDB Not Found

ModuleNotFoundError: No module named 'chromadb'

Solution: Reinstall dependencies

uv sync

MongoDB Connection Error

❌ Error: MONGODB_URI not properly set in .env file

Solution: Add your MongoDB Atlas connection string to .env with actual password

MONGODB_URI=mongodb+srv://username:YOUR_PASSWORD@cluster0.mongodb.net/...
MONGO_DB_NAME=rab_test
MONGO_COLLECTION=rab_test

---

📚 Additional Resources

• LangChain Docs: https://python.langchain.com/
• LangGraph Docs: https://langchain-ai.github.io/langgraph/
• ChromaDB Docs: https://docs.trychroma.com/
• MongoDB Atlas: https://www.mongodb.com/atlas
• LangChain-MongoDB: https://github.com/langchain-ai/langchain-mongodb
• Ollama Models: https://ollama.ai/library
• Tavily Search: https://tavily.com/

---

🤝 Contributing

Feel free to add more examples or improve existing ones!

📝 License

MIT

---

🛠️ Development

Run with uv

# Run any script
uv run python <script-name>.py

# Install new dependency
uv add <package-name>

# Sync dependencies
uv sync

IDE Configuration (PyCharm)

1. Set Python interpreter to .venv/bin/python
2. Use "uv run" for running scripts
3. Or configure run configuration to use the virtual environment

---

📚 Key Concepts

What is LangChain?

LangChain is a framework for building applications with Large Language Models (LLMs). It provides:

• Tool calling capabilities
• Prompt templates
• Memory management
• Agent workflows

What are Tools?

Tools are functions that AI agents can call to:

• Search the web (Tavily)
• Query databases
• Perform calculations
• Access external APIs

What is Structured Output?

Using Pydantic models to get consistent, typed responses from LLMs:

class AgentResponse(BaseModel):
    answer: str
    sources: List[Source]

Local vs Cloud LLMs

Feature	Ollama (Local)	OpenAI (Cloud)
Cost	Free	Pay per token
Privacy	Runs locally	Sends to API
Speed	Fast (no network)	Depends on network
Quality	Good	Better for complex tasks

---

🐛 Troubleshooting

"ModuleNotFoundError: No module named 'langchain_community'"

uv add langchain-community

"TAVILY_API_KEY not found"

Make sure .env file exists in project root with:

TAVILY_API_KEY=tvly-dev-your-actual-key-here

"Connection error" with Ollama

1. Make sure Ollama is running:

   ollama serve

2. Check if model is installed:

   ollama list
   ollama pull llama3.1:8b

Run button in IDE uses wrong Python

Configure PyCharm to use .venv/bin/python or run with uv run command.

---

📖 Learn More

• LangChain Documentation
• Tavily API Docs
• Ollama Documentation
• Pydantic Documentation

---

🎓 Learning Path

1. Start here: main_2summerization_with_local_ollama.py - Understand basic LangChain usage
2. Learn tools: main_3tool_call.py - See how agents call functions
3. Add search: main_5simple_tavily.py - Real-time web search
4. Advanced: main_6job_search_example.py - Build practical agents

---

🤝 Contributing

Feel free to:

• Add more examples
• Improve documentation
• Report issues
• Suggest features

---

📝 License

MIT License

---

👨‍💻 Author

Ruby Learner Myanmar

---

Happy Coding! 🚀

Last updated: December 29, 2025