☕ LangGraph Starbucks AI Agent

A production-grade Conversational AI Agent that allows users to order Starbucks drinks through natural, multi-turn conversations.

Built using NestJS, LangChain LangGraph, Google Gemini, and MongoDB, this project demonstrates how to design stateful AI agents with memory, tool execution, and schema-driven reasoning.

---

🧠 Overview

This AI agent behaves like a smart Starbucks barista.
It understands user intent, asks follow-up questions, validates structured order data, confirms the order, and finally persists it in a database — all while maintaining conversational context across multiple turns.

The system is designed using LangGraph, enabling fine-grained control over agent workflows, decision-making, and tool invocation.

---

🚀 Key Features

• 🧩 Conversational AI Agent for Starbucks drink ordering
• 📦 LangGraph-powered state machine for controlled multi-step conversations
• 🛠️ Tool calling to create and manage orders in a real database
• 🗄️ MongoDB persistence with LangGraph checkpointers for memory
• ✅ Zod schema validation to ensure reliable, structured AI outputs
• 🔄 Order confirmation and progress tracking
• 😀 Human-like, friendly responses with emojis and contextual suggestions
• 🔐 Authentication & User Sessions
  • Secure user identification for conversations
  • Conversation memory scoped per authenticated user
  • Prevents cross-user data leakage in multi-session environments

---

🏗️ Architecture Highlights (HR-Ready)

• Stateful AI Agent Design using LangGraph annotations and reducers
• Schema-driven AI workflows to prevent hallucinated or malformed outputs
• Tool-based execution model where the LLM triggers real backend actions
• Checkpointed memory allowing conversations to resume seamlessly
• Separation of concerns between AI logic, validation, persistence, and APIs

This mirrors how modern AI systems are built in production environments.

---

🛠️ How It Works

1. User sends a message (e.g., "I want a latte with oat milk")
2. The AI agent checks which order details are missing
3. The agent asks follow-up questions contextually
4. Once all fields are complete, the agent asks for confirmation
5. On confirmation, the create_order tool is called
6. Order is validated and saved to MongoDB
7. The response always includes structured JSON for frontend use

---

⚙️ Setup & Installation

1️⃣ Clone the Repository

git clone https://github.com/yourusername/langgraph-starbucks-agent.git
cd langgraph-starbucks-agent

2️⃣ Install Dependencies

yarn install
# or (Windows)
npm install

3️⃣ Configure Environment Variables

Create a .env file in the root directory:

GOOGLE_API_KEY=your_google_api_key
MONGO_URI=mongodb+srv://username:password@cluster.mongodb.net/starbucks_db

4️⃣ Run the Server

yarn start:dev
# or (Windows)
npm run dev

---

🧪 Example Conversation

User: I want a caramel macchiato
AI: What size would you like? ☕
User: Grande
AI: Would you like any toppings? 🍫
User: Yes, whipped cream
AI: Please confirm your order: Grande Caramel Macchiato with Whipped Cream ✅
User: Yes
AI: Order created successfully 🎉

---

🗄️ Database Details

• Database: drinks_db
• Collections:
  • orders – saved drink orders
  • langgraph_checkpoints – conversation memory

All orders are validated using Zod schemas before insertion.

---

🧭 Upcoming Enhancements

1️⃣ Domain Expansion

• 🏦 Banking Assistant (accounts, cards, loans)
• 🏥 Hospital Appointment Agent
• 💼 SaaS Customer Support Agent

2️⃣ RAG Integration

• Knowledge-base powered responses using vector search

---

📜 License

MIT License