Steps to Create Intelligent Scenarios

Open the browser and navigate to the Fiori Launchpad by clicking here. We've already kept the system logged in. And under "Analytics" tab launch the "Intelligent Scenarios" application.
Click the Create button and choose Side-by-Side.
Next, go to the Settings tab and select Generative AI as the Scenario Type.
Enter unique Scenario name in the below format, where ## is your attendee ID.
```
Z_POL_DOC_SUMM_##
```
Enter the description
```
Summarize Return Policy Document
```
Select Stateless – Customer as the Usage Type.
This option provide the shared connectivity model allows a single connection between the ABAP system and the SAP Generative AI Hub to be configured once and reused across multiple AI use cases. This simplifies administration, improves governance, and reduces setup effort. Reuse connectivity is already set up and no action is required here for connectivity.
Add a Generative AI model to the Intelligent Scenario by selecting Add Model and providing the required details.
Enter the below details:
Model name as Z_POL_DOC_SUMMARIZE_MOD
Description as Model with orchestration modules
Select the appropriate Executable ID from the drop down. In this case, we will choose azure-openai
Select the corresponding Model from the drop down. In this exercise, we will choose gpt-5-mini
Optionally, choose a specific Model Version (latest is selected by default). In this case, we can keep as empty to have latest.
Choose Add to confirm
Intelligent scenario with Orchestration service:
Navigate to the Execution Flow Template
Upload the Execution Flow Template
Click the Upload button.
Locate the Orchestration config folder on your desktop.
Open the config_json.txt file to view the orchestration configuration JSON.

Either select all and copy the JSON, or paste the below JSON directly into the dialog.

{
  "module_configurations": {
    "grounding_module_config": {
      "type": "document_grounding_service",
      "config": {
        "filters": [
          {
            "id": "filter1",
            "data_repositories": [
              "2a35766b-e72e-4ef8-9223-5995c839b1cd"
            ],
            "search_config": {
              "max_document_count": 0
            },
            "data_repository_type": "vector"
          }
        ],
        "input_params": [
          "grounding_input_variable_1"
        ],
        "output_param": "grounding_output_variable"
      }
    },
    "llm_module_config": {
      "model_name": "gpt-5-mini",
      "model_params": {},
      "model_version": {}
    },
    "templating_module_config": {
      "template_ref": {}
    },
    "filtering_module_config": {
      "input": {
        "filters": [
          {
            "type": "azure_content_safety",
            "config": {
              "Hate": 0,
              "SelfHarm": 0,
              "Sexual": 0,
              "Violence": 0
            }
          },
          {
            "type": "llama_guard_3_8b",
            "config": {
              "child_exploitation": true,
              "code_interpreter_abuse": true,
              "defamation": true,
              "elections": true,
              "hate": true,
              "indiscriminate_weapons": true,
              "intellectual_property": true,
              "non_violent_crimes": true,
              "privacy": true,
              "self_harm": true,
              "sex_crimes": true,
              "sexual_content": true,
              "specialized_advice": true,
              "violent_crimes": true
            }
          }
        ]
      },
      "output": {
        "filters": [
          {
            "type": "azure_content_safety",
            "config": {
              "Hate": 0,
              "SelfHarm": 2,
              "Sexual": 4,
              "Violence": 6
            }
          },
          {
            "type": "llama_guard_3_8b",
            "config": {
              "child_exploitation": true,
              "code_interpreter_abuse": true,
              "defamation": true,
              "elections": true,
              "hate": true,
              "indiscriminate_weapons": true,
              "intellectual_property": true,
              "non_violent_crimes": true,
              "privacy": true,
              "self_harm": true,
              "sex_crimes": true,
              "sexual_content": true,
              "specialized_advice": true,
              "violent_crimes": true
            }
          }
        ]
      }
    },
    "masking_module_config": {
      "masking_providers": [
        {
          "type": "sap_data_privacy_integration",
          "method": "anonymization",
          "entities": [
            {
              "type": "profile-email"
            },
            {
              "type": "profile-gender"
            },
            {
              "type": "profile-pronouns-gender"
            },
            {
              "type": "profile-location"
            },
            {
              "type": "profile-person"
            },
            {
              "type": "profile-url"
            }
          ],
          "mask_grounding_input": {
            "enabled": true
          },
          "allowlist": [
            "ANALYTICS",
            "SPECIALIST"
          ]
        }
      ]
    },
    "input_translation_module_config": {},
    "output_translation_module_config": {}
  }
}

Once a valid JSON is uploaded, the modules will be displayed with their parameters and values in display mode.
In the Grounding subsection, add the required data repository type Vector. This will support the grounding document addition.

To restrict your grounding module output during inference, maintain Search Configuration value. This can help to retrieve only relevant data instead of all the uploaded data.
The Input Translation module allows you to translate LLM text prompts and grounding module output into a target language. It may help improve LLM response when the configured model performs better with input in specific language, example English. In this use case, prompt texts are already English, so no need to maintain the Input translation configuration.
Also, Prompt Shield is already true as per the JSON configuration file.
The Data Masking module enables the anonymization or pseudonymization of data before it is sent to the LLM model for processing.
The Input filters defined in the Execution Flow Template JSON are populated and displayed in the table.
The Output Filtering module allows you to filter the harmful or hateful content generated by the LLM.
The Output Translation module allows you to translate LLM response into a target language. It helps to display the LLM response in the language user logged, if the translation is supported.
Add Prompt & Grounding Template
Add a Grounding Template with name GROUNDING_TEMPLATE with description Grounding query and then add the below Grounding template text.
```
{ISLM_GROUNDING_QUERY}
```
Select the Display template information as Yes. The grounding template can include dynamic parameters similar to the user prompt.
The Grounding Template retrieves relevant information from the selected data repository and appends it to the user prompt. It supports dynamic parameters, which are provided at runtime (during inference).
Add Prompt template for System prompt
Enter the prompt name SYSTEM_PROMPT. The system prompt is used to set the overall context, behavior, or persona for the AI's responses. It provides the fundamental instruction set to guide the model's behaviour throughout an interaction.
Enter the description Explain the role and responsibility for the LLM
Select the Display template information as Yes.
Enter the below text as Prompt text: You are {ISLM_ROLE}. Your responsibility is to {ISLM_RESPONSIBILITY}.
Default parameter value of ISLM_ROLE: An helpful assistant.
Default parameter value of ISLM_RESPONSIBILITY: Assist the user with their queries.
Add Prompt template for User prompt
Enter the prompt name: SUMMARIZE_RETURN_POLICY
Enter the description Use LLM to summarize the return policy document to populate standard JSON
Select the Display template information as Yes.

Enter the below text as Prompt text

Generate a summary of return policy document provided below for the company {ISLM_COMPANY_NAME}.

### INSTRUCTIONS
- Populate the conditions of the return policy in the following JSON format
{
  "return_policy": {
    "overview": {
      "summary": "string",
      "free_returns": "boolean",
      "customer_friendly_notes": "string"
    },
    "return_window": {
      "days": "integer",
      "from_date": "string", // e.g., "from delivery date" or "from purchase date"
      "extensions": ["string"] // e.g., ["holidays", "defective items"]
    },
    "eligibility_criteria": {
      "conditions": ["string"], // e.g., ["unused", "original packaging", "with tags"]
      "proof_of_purchase": {
        "required": "boolean",
        "alternatives": ["string"] // e.g., ["order history", "email confirmation"]
      },
      "item_condition": "string" // e.g., "unworn, unwashed"
    },
    "non_returnable_items": {
      "categories": ["string"], // e.g., ["software", "undergarments", "perishables", "custom products", "final sale"]
      "reasons": ["string"]
    },
    "return_process": {
      "initiation_methods": ["string"], // e.g., ["online portal", "email support", "phone", "in-store"]
      "steps": ["string"],
      "return_authorization": {
        "required": "boolean",
        "form_url": "string",
        "label_provided": "boolean"
      },
      "shipping": {
        "who_pays": "string", // e.g., "customer", "merchant", "prepaid label"
        "instructions": "string",
        "address": "string",
        "preferred_carriers": ["string"]
      },
      "required_documents": ["string"] // e.g., ["receipt", "order number", "packing slip"]
    },
    "refund_options": {
      "types": ["string"], // e.g., ["full refund", "partial refund", "store credit", "exchange"]
      "default_method": "string",
      "payment_method": "string", // e.g., "original payment", "gift card"
      "processing_time": "string", // e.g., "3-5 business days"
      "fees": {
        "restocking_fee": {
          "amount": "number",
          "percentage": "number",
          "applies_to": ["string"]
        },
        "other_fees": ["string"]
      }
    },
    "shipping_costs": {
      "original_shipping": "boolean", // refundable?
      "return_shipping": "boolean" // refundable?
    },
    "special_cases": {
      "defective_damaged": {
        "extended_window": "boolean",
        "free_shipping": "boolean",
        "full_refund": "boolean"
      },
      "wrong_item": {
        "process": "string"
      },
      "holidays": "boolean",
      "other": ["string"]
    },
    "contact_info": {
      "support_email": "string",
      "phone": "string",
      "live_chat": "boolean",
      "return_center_url": "string"
    },
    "legal_notes": {
      "state_laws": ["string"],
      "warranties": "string"
    },
    "last_updated": "string" // for policy version tracking
  }
}

- Respond with only the JSON output

### RETURN POLICY DOCUMENT
{ISLM_GROUNDING_OUTPUT}

If you have configured grounding module, at least one User Prompt must be added with the following parameter: {ISLM_GROUNDING_OUTPUT}
At runtime, this placeholder is automatically replaced with the content retrieved by the Grounding Template from the configured data repository.
Save Draft
Navigate to Scenario documents section by pressing Back
Upload grounding documents to an Intelligent Scenario
Press Upload button
Select the below files from the Grounding files folder in your desktop
Enter the document names like below and choose English as language for all the files.
Click Publish button and enter the package as local object $TMP
Press OK

Intelligent Scenario is created successfully.