README.md

Exercise 2 - Prompt Templating

In this exercise, you will explore how to create and use client-side prompt templates to dynamically generate and send prompts. This will enable you to streamline user interactions and tailor responses efficiently within your application.

1. Navigate to the Function

Open orchestration.ts file and search for the function orchestrationCompletionTemplate.

2. Add Implementation

Type or uncomment the following code in the function orchestrationCompletionTemplate:

const orchestrationClient = new OrchestrationClient({
    llm: {
        model_name: 'gemini-1.5-flash',
        model_params: { max_tokens: 1000, temperature: 0.1 }
    },
    templating: {
        template: [
            { role: 'system', content: 'Please generate contents with HTML tags.' },
            {
                role: 'user',
                content: 'Create a job post for the position: {{?position}}.'
            }
        ]
    }
});

const response = await orchestrationClient.chatCompletion({
    inputParams: { position: 'Java dev' }
});

return response.getContent();

Note

In this exercise, a few notable modifications are introduced to improve the model’s flexibility and input handling:

• You’ll configure an additional model option, temperature, which controls the randomness of the model’s responses.
• A system prompt will be added to guide the model’s behavior.
• You will use a client-side prompt template that includes a placeholder, position.
• When calling the chat completion endpoint, you’ll pass an input parameter to provide the value of the position variable.

3. Restart the Application

Save your changes and wait for the application to restart automatically.

4. Check the LLM Response

Open your browser and visit http://localhost:8080/orchestration/template. You should see a nice HTML page generated by the LLM.

Important

We strongly recommend adjusting the following values to meet your specific use case:

• Model options
• System prompt
• User provided prompt (template)
• Parameters for populating variables defined in the prompt template

Summary

Excellent!

Now, let’s take a closer look at the key concepts you’ve learned so far.

• Client-Side Prompt Templates: You demonstrated how to use prompt templates on the client side to send prompts dynamically.
• System & User Prompts: In addition to handling user-provided prompts, you also showcased how to send an instruction through a system prompt, enabling more control over the response behavior.

You’re now ready to move on to the next step. Continue to Exercise 3 - Content Filtering