Sample prompts

Edit online
Draft comment:
This topic is shared by BAW, CP4BA, CP4BASaaS. Last updated on 2025-03-13 12:15
Explore these sample prompts to understand how Large Language Models (LLMs) can be utilized in your service flow.

To get the most out of these examples, you can experiment with different LLMs and adjust the values in the Parameters section. This will allow you to see how the generated output changes based on your inputs.

Table 1. Sample prompts
Example Context Input
Generate loan acceptance letter Generate an acceptance letter to the loan request provided as input.

Customer:{{customerName}}

Reference:{{ref}}

Amount:{{amount}}
Generate translation The input may be in another language, translate it into English {{message}}
Generate sentiment Analyze the sentiment of a customer review provided as input and give it a note between 0 and 5 included, where 0 means "very unsatisfied" and 5 means "very satisfied". The output must only be made of a single digit. {{customerReview}}

Sample prompts for text-only LLMs

For LLMs that only support text prompts, the prompt can be enhanced with string variables and by attaching plain text documents. The content of the plain text document will be appended to the text prompt.

The Gen AI page for a meta-llama/llama-3-3-70b-instruct model shows the following options: Use the attached document as a template when creating a response in the Prompt -> Context area, Tell me about {{topic}} in the Prompt -> Input area, and winter tires as a value in the Variables -> Value column.

When you examine the RAW prompt, the content of the test data file is not included but is replaced with information about the size and type of the file.

RAW prompt for a meta-llama/llama-3-3-70b-instruct model shows that the specific prompt template is used.

Sample prompts with a training example

When a training example is saved from the generated response, the test data is included in the training example. For document attachments, this makes it possible to include a static server file document in the prompt without including it as a variable in the service flow by saving the training example, then removing the attached document variable from the prompt. If you need to switch to a different server file, a new training example must be created.

The Gen AI page shows a training example with the prompt Tell me about winter tires, including the attached file response Template.txt as test data, and an expected output that describes winter tires and their benefits.

When you examine the RAW prompt in this scenario, you notice that the training example includes the document but there is no document in the new prompt content. The test data value for the topic variable was changed from winter tires to bananas before obtaining this RAW prompt.

The RAW prompt for the training example doesn't include a document in the prompt content.

Sample prompts with binary documents

For LLMs that support binary documents such as images, audio, or PDF, the content for each attached document variable is included in the multimodal prompt for the LLM.

For the meta-llama/llama-4-maverick-17b-128e-instruct-fp8 LLM that supports binary documents such as images, the Gen AI page shows a sample prompt where attached image files are included in the multimodal input to the LLM.

When you examine the RAW prompt, the base64-encoded content is replaced with the size and type of file.

RAW prompt for a model that supports binary documents.