Build an AI‑Powered Airline Ticket Agent with Spring AI Alibaba

0 Introduction

This article explains how to use Spring AI Alibaba to build an AI Agent application for an intelligent airline‑ticket customer‑service platform.

1 Requirements

Understand user intent through a large‑model conversation.

Support multi‑turn dialogue with context awareness.

Comply with airline regulations, ticket‑change and cancellation rules.

Invoke auxiliary tools when necessary.

2 Technical Architecture

2.1 Integrating the Large Model

The application is a standard Spring Boot service that receives user queries, forwards them to an AI model, and lets the model guide the business flow. The simplified architecture shows the model acting as an agent that interprets user requests and decides the next action.

2.2 Using RAG for Ticket Rules

Pure reliance on a generic LLM cannot guarantee correct handling of airline‑specific policies. To provide reliable answers, the article introduces Retrieval‑Augmented Generation (RAG) to inject domain knowledge—such as airline change‑fee rules—into the model.

How does the model know whether a user meets the refund criteria?

How does it understand the specific change‑fee regulations for each airline?

By loading the ticket‑policy documents into a vector store and attaching them to the prompt, the model can reference the exact rules before making a decision.

With RAG, the agent behaves like a well‑trained customer‑service representative that can both converse naturally and enforce airline policies.

2.3 Function Calling for Business Actions

While the AI decides what should happen, the actual execution—such as updating a booking record—must be performed by the application. Spring AI’s function‑calling feature maps model decisions to concrete Java methods (e.g., getBookingDetails, changeBooking, cancelBooking) that modify the database.

2.4 Chat Memory for Multi‑turn Dialogue

Large models are stateless, so to support multi‑turn conversations the application must retain previous exchanges. Spring AI Alibaba provides a built‑in Conversation Memory component that automatically stores and injects the dialogue history into each new prompt.

3 Core Capabilities of Spring AI Alibaba

Chat Model API for basic conversational ability.

Prompt management utilities.

Chat Memory for multi‑turn context.

RAG and Vector Store integration for domain‑specific knowledge (e.g., ticket policies).

4 Implementing the Agent with ChatClient

Spring AI Alibaba offers a high‑level ChatClient API that composes the above capabilities (memory, RAG, function calling) into a fluent, stream‑based workflow.

this.chatClient = modelBuilder
        .defaultSystem("""
            您是“Funnair”航空公司的客户聊天支持代理。请以友好、乐于助人且愉快的方式来回复。
            您正在通过在线聊天系统与客户互动。
            在提供有关预订或取消预订的信息之前，您必须始终从用户处获取以下信息：预订号、客户姓名。
            在询问用户之前，请检查消息历史记录以获取此信息。
            在更改预订之前，您必须确保条款允许这样做。
            如果更改需要收费，您必须在继续之前征得用户同意。
            使用提供的功能获取预订详细信息、更改预订和取消预订。
            如果需要，可以调用相应函数完成辅助动作。
            请讲中文。
            今天的日期是 {current_date}.
          """)
        .defaultAdvisors(
            new PromptChatMemoryAdvisor(chatMemory), // Chat Memory
            new VectorStoreChatMemoryAdvisor(vectorStore)),
            new QuestionAnswerAdvisor(vectorStore, SearchRequest.defaults()), // RAG
            new LoggingAdvisor())
        .defaultFunctions("getBookingDetails", "changeBooking", "cancelBooking") // FUNCTION CALLING
        .build();

Injecting this ChatClient as a regular Spring bean enables the ticketing service to gain AI‑driven capabilities without dealing with low‑level model calls.

5 Running Result