AI Companion

Introduction

The AI app gives you the opportunity to enrich or translate data using ChatGPT. This App sets a new standard with its unique ability to use downloadable links as a source attribute. Simply upload or link to a file, and the app extracts and enriches product data automatically. Effortlessly generate detailed drafts, enrich attributes from external sources, and translate content into multiple languages—all in one streamlined platform. Perfect for scaling global operations with ease.

Connection

Go to the Settings icon (Top left) and choose Connections. Use the + CREATE button and select Type Ai connection LLM, fill in the rest of the requirements. Once created, insert your API-key and select your preferred model. Click TEST connection to validate, and save to store your credentials to the connection.

Product translation

Goto the App Tasks section, use the + CREATE button to create a new Task and choose, ai_translate_product as type. Once created, you see Parameters options available for you.

• First parameter is Ai Connection (required). Here you select the connection you made earlier.

• Second parameter is Channel (required). This determines the Akeneo channel (if the attribute value is scopable by channel).

• Next parameters are Source and Destination Locale. These options determine what your source and target languages are.

• Last important parameter is the Attribute Selection. This determines the attributes you wish to translate.

Here is a small example where we translate a description from French to German.

connection: 'openai-gpt4o'

channel: 'ecommerce'

source_locale: 'fr_FR'

target_locale: 'de_DE'

attributes: ['description']

Once saved, you can RUN the task, and it will create an operation that performs your task.

Catalog translation

Goto the App Tasks section, use the + CREATE button to create a new Task and choose, ai_translate_catalog as type. Once created, you see Parameters options available for you.

• First parameter is Ai Connection (required). Here you select the connection you made earlier.

• Next parameters are Source and Destination Locale. These options determine what your source and target languages are.

• Last important parameter is the Catalog Type Selection. This determines the entity labels you wish to translate.

Here is a small example where we translate attribute labels from French to German.

connection: 'openai-gpt4o-mini'

source_locale: 'fr_FR'

target_locale: 'de_DE'

entity: 'attribute'

Once saved, you can RUN the task, and it will create an operation that performs your task.

Product completion

Goto the App Tasks section, use the + CREATE button to create a new Task and choose, ai_autocomplete_product as type. Once created, you see Parameters options available for you.

• First parameter is Ai Connection (required). Here you select the connection you made earlier.

• Second parameter is Channel (required). This determines the Akeneo channel (if the attribute value is scopable by channel).

• Next parameter is Source Locale. This determines the Akeneo locale (if the attribute value is scopable by locale).

• The final parameters are checkboxes that show more grouped parameters per autocompletion.
  
  

! Make sure the locale you provide has accurate labels, as it will help the LLM understanding the reques.

The AI component

The Ai_Component is an attribute that is automatically generated on your Pim  when you run the Task ai_autocomplete_product for the first time. The Task will look for products where the Ai_Component is set to auto_complete. By default we don’t set a value. In order to make your product visible for Ai App, you will have to do the following:

1. SET the Ai_Component attribute in your family(ies).

2. SET the Ai_Component to auto_complete on the products you wish to be visible.

Once saved, you can RUN the task, and it will create an operation that performs your task.

Once the product is completed by the AI, it will SET the Ai_Component to ready_to_review.

[CheckBox] Parameter Autocomplete Product

This checkbox will help you enrich empty values of a given product. There are a couple of possible choices to further enrich your product values:

• based on it’s own content:
  Let’s say you have a product that is 80% ready but still misses some required attributes. Maybe you added some metric attributes recently that are available inside the product description. In that case your product will provide enough context to further enrich the product data. You don’t need to select a parsable attribute.

• based on source content:
  Let’s say you have a product that is NEW and you have a PDF document, or a PDF link that contains useful source content to further enrich the product data. In this case the product itself has not enough context, but you can provide a parsable attribute with the correct product information. A parsable attribute can be a media-attribute with a PDF document, or a text-attribute with an external link to a PDF document.

• based on Exclude / Include attributes:
  These parameters determine the attributes you wish to exclude or include as enrichable product values. We support the following Types: Metric, Options, Text, TextArea, Boolean and Date.

[CheckBox] Parameter Autocomplete Product Categories

This checkbox will help you enrich the correct category for a given product. You have the option to narrow the category selection by selecting a root category. The option Top level categories only will narrow the category list to only include top level categories. The Category instructions parameter are the prompt instructions passed to the LLM.

[CheckBox] Parameter Autocomplete Product Family

This checkbox will help you enrich the correct family for a given product. The Family instructions parameter are the prompt instructions passed to the LLM.

Specific prompts per Attributes

If you want to set a specific prompt for an attribute, for example “make this description 300 words long”, you can set this prompt details per attribute. In Akeneo under the attribute guidelines select the en_US locale. For example:

create a shorter description from the 'description' attribute

Best practices

• Experiment with different models and test out what models works best for you.

• Enrich your catalog first, make sure your catalog data is labelled correctly and your entity codes are meaningful.

• Experiment with prompts and tailor the prompts to you liking.