Get Started with Generative AI
This guide will help you leverage Generative AI capabilities of ONE to automate and enhance your data quality and management tasks. It includes practical examples and tips for proper prompt engineering, ensuring a smooth experience as you explore the potential of AI in your workflows.
Dive into the platform and experiment with these powerful features to unlock new efficiencies in your data quality processes.
|
How to use the guide?
We recommend reading through this article before using Generative AI in ONE as it sets the foundation for exploring Generative AI features of ONE. |
Generative AI in ONE
ONE offers two main options for incorporating Generative AI:
- Embedded features
-
Tools integrated directly into the platform, such as text-to-SQL, text-to-DQ-rule, and generating descriptions, which are available in Ataccama Cloud environments.
- AI Agent
-
A powerful assistant capable of handling both atomic tasks and complex multi-step workflows. This option is not yet available in the product.
Key embedded features
The following are some of the key currently available embedded Generative AI capabilities:
-
Text-to-SQL/SQL-to-Text: Write plain-language prompts to generate SQL queries.
-
Example: “Generate a query to select customer names and emails where the credit limit exceeds $100,000, ordered by credit limit.”
-
Output:
SELECT customername, email FROM customers WHERE creditlimit > 100000 ORDER BY creditlimit DESC;.
-
-
Text-to-DQ Rule: Create and debug data quality rules from natural language descriptions.
-
Example: “Create a DQ rule to validate email addresses.”
-
-
Enhanced Descriptions: Automatically generate or improve descriptions for catalog items, terms, and attributes.
-
Example: "Generate a description for the column 'customer_id' in the 'orders' table."
-
-
Explain ONE Expression: Simplify complex expressions by converting them to plain English.
-
UI Translation and Text Tools: Localize user interfaces and refine text clarity.
|
Recommended resources
|
Tips for prompt engineering
It’s common to underestimate how much guidance AI needs to provide useful and accurate answers. To get the most out of Gen AI, consider the following best practices for crafting prompts you could submit to Gen AI features like Data Quality Rule Generation, Chat with Documentation, or Text-to-SQL:
- Treat the AI like a human intern
-
When you give the AI a prompt, ask yourself whether a human intern would be able to perform the task with the same information. In other words, treat the AI like someone who might need extra context to be able to accurately perform the task at hand.
- Be specific
-
Be specific about the goal of your prompt and make sure to define the details in the prompt clearly. Let’s say you want to create a DQ rule to “find data records in Q2”. It might be helpful to define what “Q2” means. For example: ”Find data records that were logged in April, May, or June.”
-
Ineffective prompt example: “Tell me about the data.”
-
Effective prompt example: “List all tables containing customer information in the catalog.”
-
- Use context
-
Add context to the prompt to make sure you fill in any missing potential details.
-
Ineffective prompt example: “Find null values.”
-
Effective prompt example: “Find columns with more than 50% null values in the sales dataset.”
-
- Provide examples in the prompt
-
When asking for rules or descriptions, it can be helpful to include sample data or expected formats.
-
Ineffective: “Check for SSNs.”
-
Effective: “Check for US Social Security numbers that have the format of 123-45-6789.”
-
- Iterate
-
Start with a general prompt and refine based on the output. For example:
-
Initial prompt: “Generate a DQ rule for emails.”
-
Iteration: “The rule should validate that emails include a domain and an '@' symbol.”
-
FAQ
The following are some common questions and concerns:
-
Is customer data used to train the models?
No, Ataccama does not use customer data or metadata for model training.
-
Can features be turned off?
Yes, administrators can turn off AI features in Global Settings.
-
What model powers Ataccama’s AI?
Generative AI features are powered by OpenAI’s GPT-4 on Azure AI.
-
What if I encounter issues?
Your feedback is invaluable. Report any bugs or suggest improvements through our support channels.
Was this page useful?