[Season 1] Spring AI in Action: Developing a Premium Reservation AI Agent

🏆 Goal

1. Effective Agent Design: Implement a robust system with separated brain-memory-tools as shown in the architecture diagram above.

2. Tool Calling Master: Make AI autonomously judge and call Java methods (business logic) to manipulate databases.

3. Business Logic Integration: Rather than simple Q&A, we build a complete backend system where Read (R) - Create (C) - Delete (D) operations are organically connected.

🎩 Concept

"GourmetBot" - AI Concierge for Premium Fine Dining

• Smart: Remembers customer visit history and suggests personalized tables.

• Safe: Essential allergy information is checked and data integrity is guaranteed.

• Polite: Uses an elegant and formal tone and manner that is not mechanical.

  
  

🔑 Core Content

1. Spring AI Architecture Implementation: Implementing each element of the diagram in code (System Prompt, Repository, Tools).

2. Advanced Prompt Engineering: Applying Slot Filling (Follow-up Questions), Time Awareness (Date Calculation), and Chain of Thought (Step-by-Step Reasoning) Techniques.

3. Data Integrity & Transactions: Implementing Upsert Pattern (automatic user registration) and Soft Delete (reservation cancellation status management).

4. Tool Strictness: Powerful input data validation using Java Record (DTO) and JSON Schema.

(Season 1) AI Agent Architecture

1. Central Control (Spring AI & LLM): The framework receives user requests and forwards them to the LLM (GPT). At this point, the System Prompt we wrote is injected to become the agent's 'judgment criteria (brain)'.

2. Hands and Feet (Tools - Java Code): LLMs cannot access databases on their own. Instead, based on their judgment results, they call (Tool Calling) the business logic (reservation creation, cancellation, inquiry) that we implemented in Java.

3. Memory Storage (Memory - MySQL): Permanently stores and retrieves conversation context (Chat Memory) and actual business data (customers, tables, reservation information).

4. Note: The grayed-out Retrieval(RAG) section is the 'knowledge search' part that will be covered in Season 2, and will not be implemented in Season 1.

   
   

(Season 1) Tool Calling

• User Request (Chat Request): The user requests a reservation from the chatbot. For example, they might request "Please make a reservation for 4 people tomorrow at 7 PM."

• Tool Definition: Spring AI defines Java methods that perform functions such as creating reservations, checking reservations, and canceling reservations as tools that AI models can understand. Each tool includes a name, description, and required input information (such as reservation date, time, and number of people).

• Dispatch Tool Call Requests: The AI model analyzes the user's reservation request and determines that it needs to call the bookTable tool to create a reservation. The AI model sends a request to Spring AI to call the bookTable tool.

• Tool Execution: Spring AI receives a request from the AI model and executes a Java method (bookTable) that performs the actual reservation creation functionality. At this time, it passes along the reservation information provided by the user (tomorrow evening 7 PM, 4 people).

• Return Result: When the bookTable method is executed and the reservation information is saved to the database, results such as reservation success status and reservation number are returned to Spring AI.

• Chat Response: The AI model provides a final response to the user based on the tool execution results. For example, it might respond, "Your reservation for 4 people tomorrow at 7 PM has been completed. Your reservation number is 12345."

(Season 1) Database ERD

• customer table: A table that stores customer information. It includes information such as customer ID, name, phone number, visit count, notes, etc.

• restaurant_table table: A table that stores restaurant table information. It includes information such as table ID, seating capacity, table type, etc.

• reservation table: A table that stores reservation information. It includes information such as reservation ID, reservation time, customer ID, table ID, number of people, allergy information, reservation status, etc. It has relationships with the customer table and the restaurant_table table.

• spring_ai_chat_memory table: A table that stores conversation content with the chatbot. It includes information such as conversation ID, conversation content, message type, timestamp, etc. Through this, the chatbot can remember previous conversation content and provide contextually appropriate responses.

🌏 Applications

This course's architecture can be directly applied to all domains that require 'reservations' and 'consultations'.

• 🏥 Hospital Appointment Bot: Return patient identification, department recommendation.

• 🏨 Hotel Concierge: Room inquiry, room service ordering.

• 💇 Beauty/Hair Salon Booking: Designer assignment, treatment time calculation.

• 🛒 CS Consultation Bot: Shipping tracking, order modification/cancellation automation.

💻 Development Environment

• IDE: IntelliJ IDEA Community Edition.

• Language: Java 17 or 21.

• Framework: Spring Boot 3.5.8 (Latest Stable).

• Library: Spring AI 1.0.3 (or 1.1.0 Snapshot).

• Database: MySQL 8.x.

• AI Model: OpenAI (gpt-4o-mini or gpt-5-mini).

📚 Learning Materials

• The source code (backend, frontend) is provided in Lecture 29 at the end of the video course.

• The course materials are provided as PDF files.

• The source code is provided through Github.

📚 Prerequisites

• Java: Understanding of basic Java syntax (Java 17+ recommended).

• Spring Boot: Basic usage of DI/IoC, JPA(Repository), Controller.

• Database: Basic understanding of SQL (SELECT, JOIN concepts).