Prompt Engineering in Spring AI

Prompt Engineering in Spring AI

In this module, we focused on Prompt Engineering — the art of guiding AI responses by carefully designing instructions. Instead of sending plain text, Spring AI gives us tools to shape prompts, reuse them, and get structured outputs that are easier to integrate in real applications.

What We Learned

• System Messages let us define the AI’s role (e.g., “act as a travel guide”), ensuring consistent tone and style.

• Assistant Messages help provide examples, so the AI learns from demonstrations and follows a consistent response format.

• Prompt Templates remove repetition by letting us create reusable prompts with placeholders for variables (like {city} or {days}).

• Externalizing Templates allows storing prompts in .st files or databases — making them easier to manage, update, and reuse without touching code.

• Structured Output lets us receive AI responses in a predictable JSON format, making it possible to map results directly into Java objects.

Why This Matters

Prompt engineering is not just about “what to ask the AI,” but how to ask it. With Spring AI:

• We can control style and behavior of the model.

• We avoid duplicating prompt strings across code.

• We make prompts maintainable and flexible, even outside of code.

• We ensure machine-friendly outputs, which can directly integrate into applications.

Example: From Plain Prompt to Structured Output

Here’s a quick reference that shows the evolution:

// Using a template with variables
PromptTemplate template = new PromptTemplate(
    "You are a travel guide. Plan a trip to {city} for {days} days."
);
 
Prompt prompt = template.create(Map.of("city", "Paris", "days", "3"));
 
TravelPlan plan = chatClient.prompt(prompt)
        .call()
        .entity(TravelPlan.class);

👉 Instead of a free-form paragraph, this returns a structured TravelPlan object with attractions and food recommendations.

Key Takeaways

• Good prompts = Better, more reliable AI responses.

• Templates + externalization = Clean, maintainable applications.

• Structured output = AI becomes directly usable in real-world apps.