Test-drive tutorial on adding AI to a Java project

[damithc]

The tutorial is at https://se-education.org/guides/tutorials/addingAiToJavaApp.html

Help needed: test-drive it and post bugs/suggestions in this thread

[FisherSkyi]

Tested on MacOS, Java zulu-17
Observation:

• Groq is easy to learn, easy to use and token limit is far from enough (use about 5k while the limit of individual model is about 500k)
• The API key is not inside the jar file, which is good for safety (if we use the suggested method putting it into .zshrc)
  For example, if we comment out the API key in .zshrc and use a fresh shell, the ai feature will not be usable with error

Caused by: dev.langchain4j.exception.AuthenticationException: {"error":{"message":"Invalid API Key","type":"invalid_request_error","code":"invalid_api_key"}}

Query (Task3)	Command (Task4)

[FisherSkyi]

Suggestion for improvement

• Add exception handle on situation where api key is not found. Possibly we can instruct user on how to set the key
• We use remote models and thus agent feature depends on internet connection to work (error w/o internet: java.net.SocketException: Network is unreachable). Would it be better if we use a local model?
• Need further instructions on how to test agent feature (Especially as output is not fully deterministic. How do we do unit test or dependency injection in this case?)

Other possible idea

• User might want the model to have some basic knowledge on time while right now the model is stateless. For example, when user input "schedule a deadline by tomorrow", the agent can transform it into real date. We might add this to system prompt:

String systemPrompt = "You are an assistant for a CLI task management app. "
    + "Today's date is " + java.time.LocalDate.now() + ".\n"
    // ... rest of the prompt

[jovnc]

Test Drive

Using MacOS, testing on iP Java Gradle project, using IntelliJ Ultimate IDE.

Content

1. At this time of testing (12/2/2026), this URL has invalid SSL certificate

https://langchain4j.dev/

Would it be better to link to the docs instead, or even their github

Code

1. Setting environment variables via the CLI doesn't work for running applications using IntelliJ IDEA. This may be a problem with students who are reliant on using IDE to run the program (which is likely common due to its convenience).

Solution: students need to manually configure the environment variables for the run (and maybe a more versatile solution like .env?)

2. AI lacks specific context to the app as system prompt may not be comprehensive enough. Eg. For my case, date needs to be format d/M/yyyy HHmm, but AI doesn't have context for this

Solution: can prompt students to do "context engineering" or modify the system prompt to suit their use cases

Other suggestions

1. Perhaps would be good to educate students to not "share" their API keys as security best practice, or even commit their API keys, which can lead to their API keys getting compromised

2. Users running the jar file are expected to have their API keys set as LLM_API_KEY to have their binaries to have the AI feature now. Would it be good to document this somewhere?

Demo

I have included my relevant code changes for others to reference (though most is generated by AI).

Related PR: jovnc/ip#10

[damithc]

Thanks all for inputs so far. I've updated https://se-education.org/guides/tutorials/addingAiToJavaApp.html to incorporate some of your inputs (the rest will be considered for future updates), and mentioned you at the bottom of the page.