LangChain with Agentic Tool Calling : AI-Powered Contact Management & Search Assistant

The purpose of this project was to test agentic tool calling based on the task assignment. This code will serve as a basis for an AI helper to take tasks and interact with the users contact list. The goal is for an intelligent AI assistant that combines contact management, email automation, and web search capabilities using Google's Gemini AI and LangChain framework. The channels for interaction with this agent will be via Telgram messages and APIs.

📋 Table of Contents

• 💡 About The Project
• ✨ Key Features
• 💻 Technologies Used
• 🛠️ Installation & Setup
• ▶️ Running Locally
• 🧪 Testing
• 📌 Project Status
• 🗺️ Future Roadmap

💡 About The Project

This project demonstrates the power of combining Large Language Models (LLMs) with structured tool calling to create an intelligent assistant capable of:

• Problem Solved: Managing contact information and automating email communications while providing intelligent web search capabilities
• Real-world Application: A virtual assistant that can look up contacts, send emails, and research information - perfect for business automation and personal productivity
• Key Learning: Mastered LangChain's agent framework and Google Gemini's function calling capabilities to create a multi-tool AI system with proper error handling and structured outputs

The assistant showcases critical thinking by validating contact information before attempting actions and problem-solving by gracefully handling failures and providing meaningful feedback to users.

✨ Key Features

• 🔍 Intelligent Contact Lookup: Search for contacts with structured validation
• 📧 Automated Email System: Send emails to found contacts with custom subjects and body content
• 🌐 Web Search Integration: Research information using Tavily search API
• 🛡️ Error Handling: Robust validation prevents actions on missing contacts
• 🧠 AI-Powered Reasoning: Google Gemini 2.5 Flash provides intelligent decision-making
• 📊 Structured Outputs: Pydantic models ensure type-safe data handling
• 🔄 Streaming Responses: Real-time agent updates for better user experience

💻 Technologies Used

Core Technologies

• Python 3.8+ - Primary programming language
• Google Gemini 2.5 Flash - Advanced LLM for AI reasoning
• LangChain - Framework for building LLM applications
• Pydantic - Data validation and settings management

APIs & Services

• Google Generative AI - Gemini model integration
• Tavily Search API - Web search functionality

Development Tools

• python-dotenv - Environment variable management
• Git - Version control system
• VS Code/PyCharm - IDE support

Key Concepts Demonstrated

• Agent-Based Architecture
• Function Calling & Tool Integration
• Structured Output Validation
• Error Handling & Graceful Failures
• Streaming AI Responses
• CRUD Operations (Contact management)
• RESTful API Integration

🛠️ Installation & Setup

Prerequisites

• Python 3.8 or higher
• pip package manager
• Google Gemini API key
• Tavily Search API key

Step-by-Step Installation

1. Clone the repository

   git clone https://github.com/yourusername/ai-contact-assistant.git
   cd ai-contact-assistant

2. Create a virtual environment

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies

   pip install -r requirements.txt

4. Set up environment variables Create a .env file in the root directory:

   GEMINI_API_KEY=your_google_gemini_api_key_here
   TAVILY_API_KEY=your_tavily_api_key_here

5. Get API Keys

   • Google Gemini API Key
   • Tavily Search API Key

▶️ Running Locally

Basic Usage

1. Activate your virtual environment

   source venv/bin/activate  # On Windows: venv\Scripts\activate

2. Run the main application

   python main.py

3. Expected Output: The agent will process the automotive repair query and display streaming responses from the AI assistant.

Available Test Cases

The application includes several test scenarios:

• Automotive Repair Query: Research brake replacement for Ford Fusion 2012
• Contact Lookup: Find and email contacts (Alice/Bob)
• Error Handling: Demonstrates graceful failure for unknown contacts

Custom Usage

Modify the inputs_success variable in main.py to test your own queries:

inputs_success = {"messages": [{"role": "user", "content": "Your custom query here"}]}

🧪 Testing

Running Tests

Currently, the project uses manual testing through the main execution block. Future versions will include:

# Planned test commands
python -m pytest tests/
python -m unittest discover tests/

Test Framework Integration

• Unit Testing: Planned with pytest
• Integration Testing: Agent workflow validation
• API Testing: Tool function verification
• Error Handling Tests: Edge case validation

Manual Testing Examples

The main.py file includes comprehensive test cases that demonstrate:

• ✅ Successful contact lookup and email sending
• ❌ Graceful handling of missing contacts
• 🔍 Complex web search queries with automotive data

📌 Project Status

🟢 Active Development - Version 1.0 Core Features Complete

• ✅ Core agent functionality implemented
• ✅ Contact lookup system with validation
• ✅ Email automation with error handling
• ✅ Web search integration
• ✅ Structured output handling

🗺️ Future Roadmap

This code will serve as a building block for other projects. The tooling structure used here will be used to build a more capable LLM integration that has long term memory for conversations. Also Telegram and API channel will be integrated for triggering access to the LLM along with structured output. The current output was to validate tool choosing based on task received.

Technologies & Resources

• Built with LangChain - Amazing LLM framework
• Powered by Google Gemini - Advanced AI model
• Search functionality by Tavily - AI-powered search

Learning Resources

• LangChain Documentation
• Google Gemini API Guide
• Pydantic Documentation

⭐ Star this repository if you found it helpful!

Built with ❤️ using Python, LangChain, and Google Gemini