AI-102 Question 415
Select 3You are tasked with creating a composed document intelligence model in Azure AI to process invoices and receipts. The invoices have consistent layouts, while the receipts vary significantly in structure. Which steps should you take to design and implement this solution effectively?
- A
Train a custom model for invoices with labeled data and add it to the composed model.
- B
Use the prebuilt Receipt model provided by Azure Form Recognizer for processing receipts.
- C
Train a custom model for receipts using labeled data instead of the prebuilt Receipt model.
- D
Combine the custom invoice model and the prebuilt Receipt model into a composed model.
- E
Deploy separate models for invoices and receipts without combining them into a composed model.
Show answer and explanation
Correct answers: A, B, D
Explanation
To create a composed document intelligence model, you should leverage custom models for documents with consistent layouts (like invoices) and prebuilt models for documents with varying structures (like receipts). Combining these models into a composed model enables a unified approach to processing multiple document types, simplifying deployment and usage.
- A. Correct.
Correct: Since invoices have consistent layouts, a custom model can be trained with labeled data and added to the composed model to handle these documents effectively.
- B. Correct.
Correct: Azure Form Recognizer provides a prebuilt Receipt model that is optimized for extracting data from receipts, which often have varying structures. This model can be used directly in the solution.
- C. Incorrect.
Incorrect: Training a custom model for receipts is unnecessary because the prebuilt Receipt model is already optimized for this purpose.
- D. Correct.
Correct: A composed model allows you to combine custom models and prebuilt models, enabling the solution to handle both invoices and receipts in a single endpoint.
- E. Incorrect.
Incorrect: Deploying separate models without combining them into a composed model would require additional logic to route documents to the correct model, which adds complexity and is unnecessary.