Transform Data
Purpose: Teach AI how to transform your data in natural language and let Transform’s AI Assistant codify it for you, no coding skill required.
Before You Begin
- Ensure you have the proper role and permissions to build inside of Transform.
Overview
Nearly every accounting task involves the reshaping of data via filters, joins, lookups, and more. Transform Data allows you to describe these data transformations to an AI assistant in natural language. The assistant then attempts to recreate your described transformation.
If it is successful and you approve of the result, the assistant will then generate custom code to ensure your Agent always behaves consistently and predictably period after period.
Inputs and the Transform Data Skill
Context is only maintained inside of individual steps of a Transform Agent. This means that you must provide the context in the form of ‘inputs’. Inputs can be added to steps manually, or you can attach the output from a previous step.
Adding Inputs Manually
Within the Transform Data skill’s preview screen:
- Navigate to the Inputs tab.
- Click Add Input.
-
Find your desired input.
- If you have previously uploaded your target input as a file, search or click the file in the Inputs tab.
- If your target input does not exist yet inside of the Agent, click the Add New button. New inputs will be added to the Input Library and will be required when running the Agent.
-
(Optional) If you selected New, you
will need to select an input type:
- File: This option allows you to upload a .csv or .xlsx data set.
- Date: This option allows you to create a Date input inside of the Agent. A Date input acts as a variable inside of the Agent so that it may update its logic period over period.
- Text: This option allows you to select a Text input inside of the Agent. A Text input acts as a variable inside of the Agent so that it may update its logic each run.
Adding Inputs from Previous Steps’ Outputs
- Navigate to the Visual Agent Builder canvas view.
- Click the small circle at the bottom of the skill block whose output you wish to serve as the input for the next step.
- Drag a line from that circle to the top of the new Skill Block.
Prompting the AI Assistant
- Navigate to the Configure tab within the Transform Data skill preview screen.
-
Craft a clear and coherent prompt
inside of the chat window.
- Remember, specificity is key here. While you may know which file, column, or cell you are looking at, the AI does not. Make sure you let your AI Assistant know exactly where you want it to work.
- (Optional)ClickEnhance Prompt to let the AI automatically polish your prompt for you.
- Press the green arrow button to submit your prompt once you are satisfied with it.
Once a prompt has been sent, the AI Assistant will read the prompt, scan the inputs, and narrate its chain of thoughts as it progresses through the request. Please note that the Assistant may ask for clarification or additional information.
When the Assistant is finished with its work, it will confirm completion and provide a new file which will be loaded into the File Preview.
Repeat this process as necessary until you are happy with the output of the current step. Keep in mind that additional messages inside of a step contribute to more and more context. Limiting steps to fewer messages can help improve AI accuracy.
Finish the Step
When you are happy with the step’s output, click on the Finish button. This button triggers the Assistant to begin codifying its work into a custom script. Please review the AI Assistant’s work prior to pressing Finish.
On the Canvas, Skill Blocks will display with a dashed border along the outside of them when they are in their draft state. Once the Skill Block has been saved and codified, the outline of the block will become solid.