🧠 AI Chatbot with Tool Calling (FastAPI + Azure OpenAI + LangChain)

This project is a production-ready AI chatbot backend built with:

✅ Azure OpenAI (GPT-4.1-Mini)
✅ LangChain Agents & Tools
✅ FastAPI backend
✅ Custom tools including email lookup & web search
✅ Extensible design to support MongoDB queries

The chatbot is capable of:

• Understanding natural language queries
• Deciding whether a tool is needed
• Calling backend functions automatically
• Returning natural human-like responses

---

🚀 Features

🤖 LLM-Powered Chatbot

Uses Azure OpenAI GPT-4.1-Mini as the reasoning engine.

🛠 Tool Calling

The chatbot intelligently calls backend functions such as:

• 🔍 web_search — Searches DuckDuckGo
• 📧 get_user_email_id — Returns email IDs based on query
• (Future) 🗄 MongoDB query tools

🌐 REST API

One simple endpoint:

POST /chat

💬 Conversational Memory

Keeps context across messages.

📜 System Prompt Control

Defines chatbot policy & behavior.

---

🏗 Architecture Overview

User → FastAPI → Azure OpenAI → Tool Selected → Backend Executes
        ↑                                       ↓
        ←────────────── Final Answer ───────────

LLM = Brain
Tools = Muscles
Backend = Orchestrator

graph TD
    A[User] --> B[FastAPI]
    B --> C[Azure OpenAI]
    C --> D{Decision Point}
    D -->|Tool Needed| E[Tool Selection]
    E --> F[Tool Execution]
    F --> G[Result Processing]
    G --> H[LLM Response]
    H --> I[Final Answer]
    D -->|No Tool| J[Direct Response]
    J --> I
    I --> A

Loading

---

📂 Project Structure

AI-chatbot-with-toolcall/
├── main.py              # FastAPI app + Agent logic
├── tools.py             # Tool functions
├── config.py            # Azure credentials
├── malay.txt            # Profile data file
├── requirements.txt     # Python dependencies
├── send_email.py        # Email sending helper
├── templates.py         # Template selection / utilities
├── email_templates/     # HTML email templates
│   ├── admin_templete.html
│   └── user_templete.html
├── frontend/            # Simple web UI for the chatbot
│   ├── index.html
│   └── README.md
└── README.md            # Backend & project docs

graph TD
    A[AI-chatbot-with-toolcall] --> B[main.py<br/>FastAPI + Agent]
    A --> C[tools.py<br/>Tool Functions]
    A --> D[config.py<br/>Azure Config]
    A --> E[malay.txt<br/>Data File]
    A --> F[requirements.txt<br/>Deps]
    A --> G[send_email.py<br/>Email Helper]
    A --> H[templates.py<br/>Template Utils]
    A --> I[email_templates/<br/>HTML Email Templates]
    A --> J[frontend/<br/>Web UI]
    A --> K[README.md<br/>Docs]

Loading

---

🌐 Frontend UI

The web UI for this chatbot is developed in a separate repository and mirrored here for convenience:

• GitHub repo: https://github.com/MalayJain412/Frontend-UI-for-AI-Chatbot-fastapi
• Local folder in this project: frontend/
• Entry file: frontend/index.html
• Frontend docs: frontend/README.md

To try it out, start the FastAPI backend, then either open frontend/index.html directly in your browser or serve the frontend/ directory with a static server (for example, VS Code Live Server or python -m http.server). Make sure any API base URL used in the frontend points to your running backend (for example, http://127.0.0.1:8000/chat).

---

⚙️ Installation & Setup

1️⃣ Clone repo & create venv

python -m venv venv
source venv/bin/activate   # Mac/Linux
venv\Scripts\activate      # Windows

---

2️⃣ Install dependencies

pip install -r requirements.txt

---

3️⃣ Configure Azure OpenAI

Create a .env file in the root directory of the project:

AZURE_OPENAI_ENDPOINT=https://YOUR-RESOURCE.openai.azure.com/
AZURE_OPENAI_API_KEY=YOUR_KEY
AZURE_DEPLOYMENT=gpt-4.1-mini
AZURE_API_VERSION=2025-01-01-preview

Note: The config.py file automatically loads these environment variables using python-dotenv. Deployment name must match your Azure Studio deployment.

---

4️⃣ Run server

uvicorn main:app --reload

Server runs at:

http://127.0.0.1:8000

---

🧰 Tools Explained

🔍 1. Web Search Tool

Searches online using DuckDuckGo API.

Used for general knowledge queries.

---

📧 2. Email Lookup Tool

Example:

get_user_email_id("What is Malay's email?")

Returns:

malayjain1234@gmail.com

Logic:

Name	Email
Malay Jain	malayjain1234@gmail.com
Aniket	anni990@gmail.com

Tool docstring tells the LLM when to use it.

---

🧠 How Tool Calling Works (Simple Explanation)

1️⃣ User asks a question
2️⃣ LLM decides whether a tool is needed
3️⃣ If yes → passes arguments to tool
4️⃣ Backend executes Python function
5️⃣ Result is returned to LLM
6️⃣ LLM writes natural reply

Example:

User: What is Malay's email?
LLM → ToolCall(get_user_email_id)
Backend returns email
LLM responds naturally

Magic ✨

sequenceDiagram
    participant U as User
    participant L as LLM
    participant T as Tool
    participant B as Backend

    U->>L: Asks question
    L->>L: Evaluates need for tool
    L->>T: Calls tool with arguments
    T->>B: Executes function
    B->>T: Returns result
    T->>L: Provides result
    L->>U: Generates natural response

Loading

---

🧩 System Prompt Role

Controls:

✔ assistant behavior
✔ tone
✔ policy
✔ general rules

Example rules:

• Prefer local tools for internal data
• Use web search only when required
• Do not mention tool names

---

🔊 Verbose Logging

verbose=True

Shows:

• tool selection
• reasoning chain
• inputs & outputs

Useful for debugging.

---

🛡 Security Notes

✔ API keys are never exposed
✔ Tools run in backend only
✔ Email lookup prevents hallucination
✔ Invalid requests return safe output

---

🧪 Testing the Chat API

Send request:

POST /chat
Content-Type: application/json

Body:

{
  "message": "What is Malay's email?"
}

---

🏛 Tech Stack

Component	Technology
Backend	FastAPI
LLM	Azure OpenAI
Agent	LangChain
Tools	Python functions
Memory	LangChain buffer

---

📌 Why LangChain?

LangChain handles:

✔ conversation history
✔ tool routing
✔ argument passing
✔ agent reasoning
✔ debug logging

So you write less glue code.

---

🧠 Future Enhancements

📌 Add MongoDB query tools
📌 Structured tool calling
📌 Auth & rate limiting
📌 Frontend UI
📌 Docker image

---

🏁 Summary

This project demonstrates a realistic production-grade AI chatbot backend that combines:

🧠 Azure OpenAI reasoning
🛠 Python business logic
🚀 FastAPI deployment
🧩 LangChain agents

It's clean, extensible, and powerful.

---

🤝 Contributing

PRs & suggestions welcome 👍

---

📄 License

MIT