top of page

Power Platform AI Builder vs. Azure Cognitive Services


Power Platform AI Builder vs. Azure Cognitive Services

AI Builder was launched in June 2019, and there were many questions about how AI Builder fair with the rest of the competition, such as Microsoft's own Azure Cognitive Services and Builder.ai.


AI Builder has been put in Preview state for a while. Now that most AI models are already in General Availability (GA), it is time to compare AI Builder in Power Platform with Azure Cognitive Services.


We can analogize AI Builder with Azure Cognitive Services by differentiating the prebuilt and custom models in AI Builder from the ones available in Azure Cognitive Services.

This is done to answer the frequent question, "why would I choose AI Builder over the already available services on Azure?"



What are Prebuilt models?

Before we get into the specifics of why and how to utilize prebuilt models in Power Apps canvas apps, let's take a moment to examine what they are about and how they vary from bespoke AI Builder models that you may train and publish.


Using AI Builder prebuilt models is like using any other out-of-the-box control or tool of the Power Platform, but with the additional benefit of using machine learning or artificial intelligence services that have already been trained based on industry data.



Custom vs. Prebuilt models

In order to make use of custom AI Builder models, you'll first need to train them and gather historical data. Manual activities, such as recognizing photos and collecting and providing important data, may be fairly thoughtful before reaching a confidence level.


The broad use of a feature has allowed Microsoft to "pre-train" some models. The Business Card Reader (BCR) component in Canvas Apps is an easy-to-understand scenario that was originally accessible in AI Builder. Behind the scenes, the BCR component uses the Form Processing model to process an image (a business card). The ability to identify an email address, a name, or a phone number does not necessitate the use of a training model. Use the component and sit back and watch the magic unfold.


Prebuilt models are based on the same design principles as the BCR component. Your Flows don't need to train the models.



GRC-114: Developing RPA Flows & AI Builder On Microsoft Power Platform


Why AI Builder?

You can leverage Azure services directly via the AI Builder toolkit when building an AI model for use in a Power Apps app or a Power Automate flow. Here are the advantages and disadvantages of each option, in my opinion:


AI Builder:

  • Advantages: Platform-level toolset, simplicity of integration

  • Disadvantages: Slower and availability depends on Microsoft’s release


Azure Cognitive Services:

  • Advantages: well-established APIs and a wide range of options

  • Disadvantages: a longer integration time is required and there are particular connectors required.



Sample Prebuilt models


Business Card Reader:

One of the characteristics of the prebuilt Business Card Reader that appeals to me is the fact that it was one of the first elements I looked at while learning about the BCR component. Imagine having a contest where guests drop their business cards into a box to be entered to win a prize at your conference booth table. Naturally, the primary goal is to collect the participants' contact information and use it in a lead/opportunity process.


If you're using a Canvas App, you'll have to scan, evaluate, and submit each card. If you have many participants, this can take a long time, but most of the scans are of a high enough quality that they don't need to be reviewed.


You might first scan the cards on your phone and then move them to a OneDrive folder using the new Flow Predict function with prebuilt models (as an example). Then use a pre-built BCR model to process the files (scanned cards) and transfer the results to your Dataverse/Dynamics lead entity via a flow triggered by the production of the files. It will be easier for me to go through the leads this time because they will be organized into batches.


Key Phrase extraction:

This is akin to the Text Analytics Key Phrases action in this instance. An important goal of this operation is extracting the most important points from a text or document. "Food" and "Great Staff" are two examples of extracts that may be obtained by feeding in the following sentence: "The food was excellent, and the staff was wonderful."


Language detection:

Here, we're focusing on the Text Analytics' Detect Language implementation. One of the primary goals is to produce a list of languages found in a text/document, which is self-explanatory.


Sentiment analysis:

In addition to Text Analytics, the Sentiment Analysis cognitive service is another prebuilt model with strong potential for automated flows. In order to use this action, you'll be given an overall emotion score between 0 and 1 (positive or negative).


Text recognition:

Now we are looking at the equivalent of the OCR-to-JSON conversion operation of Azure Cognitive Services Computer Vision. This model's goal is to identify handwritten or printed text in photographs and turn it into text that can be read by computers.



GRC-116: Developing Custom AI Models Using AI Builder


Computer Vision vs. Custom Vision

You may define the labels and train your own models to identify them using Custom Vision instead of using the Computer Vision service. A machine learning system analyzes photos as part of the Custom Vision service.


Is it possible to use computer vision APIs instead of custom vision?


Overall, the most significant distinction between the two is that the Custom Vision service can only do picture classification and object identification, as well as take in your own photos to run those functions against. More can be done with the Computer Vision APIs, but you have no say in how the models are trained. That's all I've got!


How do computer vision and machine vision vary furthermore? Automating image processing with computer vision is a long-standing practice; applying it to real-world interfaces, such as a production line, is a more recent one. Need bespoke datasets for your computer vision or machine vision project? Look no further!


As a result, what additional subjects are connected to computer vision?


Computer Vision overlaps extensively with the following fields:


Image Processing:

Image processing is concerned with the alteration of digital images. The field of Pattern Recognition focuses on the study of various methods for classifying patterns.


Photogrammetry:

Photogrammetry is the science of taking precise measurements from digital photographs. Photogrammetry and computer vision don't seem to have anything in common.


Photographic measurements are used in photogrammetry to provide more precise data. In the field of image processing, researchers look at how one picture transforms into another. In image processing, both the input and the output are images themselves. Computer vision is the production of clear, meaningful descriptions of physical things from their picture.

In summary, there are many other fields of AI services available on both Azure Cognitive Services as well as AI Builder, but it’s honey to the ear when Microsoft made them easily accessible to Power Platform and Dynamics 365 in the form of AI Builder. So now everyone can provide AI capabilities into their apps using the low code/no code way, or stick right back to Azure Cognitive Services for those pro developers with technical experiences.



FAQ


What are the capabilities of Azure cognitive services?

The cognitive services offered by Azure are they SAAS or PaaS?

Can AI Builder preview trained models for use in Power Automate without CDS access?

What is the function of Power Automate flows in AI Builder?



 

GemRain Consulting is an authorized Microsoft Partner that provides all the Microsoft certifications training. Besides official certifications training, we do provide Microsoft customized training as well. These are courses that our Master Trainer developed throughout his years of experience in training the official Microsoft courseware. He had compiled all questions and feedback commonly asked in all of his different pieces of training and crafted this 100% hands-on style workshop series.


This ensures all attendees share the same real-world experiences that our Master Trainer evangelized all these years. Most of our clients thoroughly enjoyed this variety of training and had broadly endorsed them.


GRC-116: Developing Custom AI Models Using AI Builder is our newly customized course that learners will about:


  • Understand the capabilities of Power Platform AI Builder

  • How the Power Platform AI Builder can be used to develop custom AI models to support AI workloads in your business solutions

  • How an AI model can be made available for usage in Power Apps and Power Automate flows


Commenti


bottom of page