Intent classification (check out our tutorial) is the process of using AI to categorize user input into predefined labels, allowing systems to understand the goal behind a message. This technique is fundamental for building intelligent routers that direct user queries to the appropriate business logic or service.

Example

In this example, we will use a local LLM model running via Ollama Platform. We chose to run phi3:mini-128k model because it's fast to run on local machines and uses less RAM.
Our example will show how to classify a user's inquiry. The LLM model is prompted to return only the specific enum name, which is then handled by a switch statement to simulate business routing.

package com.logicbig.example;

public enum Intent {
    GREETING,
    CHECK_ORDER,
    CANCEL_ORDER,
    REFUND_REQUEST,
    UNKNOWN
}

package com.logicbig.example;

import dev.langchain4j.model.ollama.OllamaChatModel;

public class IntentRecognitionExample {

    public static void main(String[] args) {
        OllamaChatModel model = OllamaChatModel.builder()
                                               .baseUrl("http://localhost:11434")
                                               .modelName("phi3:mini-128k")
                                               .numCtx(4096)
                                               .temperature(0.0)
                                               .build();

        String userMessage = "Where is my order?";

        String prompt = """
                You are an intent classification system.

                Classify the user message into ONE of the following intents:
                - GREETING
                - CHECK_ORDER
                - CANCEL_ORDER
                - REFUND_REQUEST
                - UNKNOWN

                Return ONLY the intent name. No explanation.

                User message:
                "%s"
                """.formatted(userMessage);

        String response = model.chat(prompt).trim();
        System.out.println("Detected intent: " + response);

        try {
            Intent intent = Intent.valueOf(response.toUpperCase());
            switch (intent) {
                case CHECK_ORDER:
                    System.out.println("start process: looking up order status...");
                    break;
                case GREETING:
                    System.out.println("start process: responding to greeting...");
                    break;
                case CANCEL_ORDER:
                    System.out.println("start process: responding to Cancel order...");
                    break;
                case REFUND_REQUEST:
                    System.out.println("start process: responding to refund request...");
                    break;
                case UNKNOWN:
                default:
                    System.out.println("start process: generic help flow...");
            }
        } catch (IllegalArgumentException e) {
            System.out.println("Process: handle unknown intent");
        }
    }
}

Output

$ mvn exec:java -Dexec.mainClass="com.logicbig.example.IntentRecognitionExample"
[INFO] Scanning for projects...
[INFO] 
[INFO] ----< com.logicbig.example:intent-classifier-lang-chain-4j-example >----
[INFO] Building intent-classifier-lang-chain-4j-example 1.0-SNAPSHOT
[INFO]   from pom.xml
[INFO] --------------------------------[ jar ]---------------------------------
[INFO] 
[INFO] --- exec:3.6.3:java (default-cli) @ intent-classifier-lang-chain-4j-example ---
Detected intent: CHECK_ORDER
start process: looking up order status...
[INFO] ------------------------------------------------------------------------
[INFO] BUILD SUCCESS
[INFO] ------------------------------------------------------------------------
[INFO] Total time:  2.431 s
[INFO] Finished at: 2026-01-14T05:28:01+08:00
[INFO] ------------------------------------------------------------------------

Conclusion

The output confirms that the LLM correctly identified the user's message 'Where is my order?' as the CHECK_ORDER intent. By constraining the model's output to specific enum values, we successfully transformed unstructured natural language into a structured programmatic flow, as evidenced by the successful execution of the corresponding switch case.

Example Project

Dependencies and Technologies Used:

• langchain4j 1.10.0 (Build LLM-powered applications in Java: chatbots, agents, RAG, and much more)
  
• langchain4j-ollama 1.10.0 (LangChain4j :: Integration :: Ollama)
  
• JDK 17
• Maven 3.9.11