Building AI and ML Applications on AWS

AI and ML have brought a paradigm shift in the way digital transformation is happening across industries. The ready availability of scalable computing capacity, massive proliferation of data, and rapid advancement in ML technologies helping customers across businesses to be part of this transformative journey. Generative applications have captured widespread imagination and attention of customers. AI and ML have been the focus of Amazon for almost 20 years and many capabilities used by Amazon customers are driven by ML.

In today’s topic we will learn about Amazon AWS capabilities of building AI and ML applications, how AI is integrated into AWS applications and we will look at some real-life case studies where AWS AI and ML is being deployed. 

AWS AI and ML  

Using artificial intelligence (AI) and machine learning (ML) on data generated by machines, devices, sensors and other systems we can optimize business operations and gain competitive advantage in business. AL/ML transform data into insight which can be used to optimize processes. AWS has comprehensive services on AI/ML. The AI/ML approach of AWS is included in three layers. 

• The bottom most layer consist of frameworks and ML infrastructure experts
• The middle layer provides ML services for developers and data scientists
• The top most layer is AI services which mimic human cognitive intelligence  

The usage of ML goes through multiple phases which involves data preparation, building and training, deploy and monitor with interface as depicted in figure below.

AWS has many products in its bucket which provide ML services such as Amazon Sagemaker, Amazon lookout for vision, Amazon lookout for equipment, AWS Panorama, Amazon Monitron etc. 

Integration: AWS AI into Applications 

AWS Sagemaker provides the capability to build, train and deploy ML models into your applications. There are hundreds of paid and free ML models available across several categories such as audio, video, computer vision, image, natural language processing, speech recognition, structured, text etc. 

AWS marketplace has pre-trained machine learning models for real time inference as well as for batch jobs. Pre-trained models do not require training data gathering, writing algorithms to train models, performing hyperparameter-optimization and training models for production. Algorithms and models can be integrated seamlessly with Amazon Sagemaker. From the Amazon management console you can use low level Amazon Sagemaker API or Amazon Sagemaker python SDK. 

How to Deploy?

To deploy a model, an AWS account requires a subscription. 

Step 1: On the listing page, click on “continue to subscribe’. Review end user license agreement and software pricing and accept offer by clicking the button

Step 2: After subscription to the listing open configuration software page for face Anonymizer. Let Fulfilment method as Amazon Sagemaker and Software Version as Version 1 as default. For region select us-east-2.
Product ARM is at the bottom of the page which is required to deploy a model using API. If you are using the Amazon Sagemaker endpoint using the console then you can ignore this.

Step 3: Choose ‘View’ in Sagemaker

Step 4: choose ‘Face Anonymizer’ listing and then select ‘Create endpoint’.

Step 5: under ‘model settings’ specify ‘face Anonymizer’ for model name and for IAM role ‘select IAM’ role which has necessary IAM permissions
 
Step 6: On ‘create endpoint’ page configure the fields as under:
Choose Face-anonymize for ‘Endpoint name’ and ‘Endpoint configuration name’
Under production variants choose ‘Edit’
In ‘Edit production variant’ box configure the following 
In instance type ‘ml.c5.xlarge) and click ‘save’

Step 7: review information and select ‘create endpoint configuration’ 

Step 8: Click on ‘Submit’ button 

AI and ML Applications: Real Life Cases

In this section we will look at some real-world examples of AWS AI/ML applications. 

• AB InBev is the largest worldwide distributor of breweries. Their delivery system reached 1 million orders in 2019 and exceeded in 2020 by month. AB InBev migrated to AWS cloud to build and deploy its solution to scale. 
• 30 MHZ provides smart agriculture solutions which provides crop farmers with real time remote crop monitoring systems to help them to optimize irrigation and ventilation, prevent diseases and sunscald, improvise pest management and predict shelf life of crops. This includes predictive models of each individual growers based on sensor data collected from greenhouse or benchmark model for new grower having no data.
• The 3M health sector company uses AWS AI to enable its health information systems while transforming clinical workflows and lab processes to help in streamlining clinical documentation and billing.
• AU Small finance bank in India has successfully onboard new customers with Video KYC from 2020. They are offering all banking services over video calls except disbursements and deposits
• Air Canada had transformed their contact centers and interactive voice response system with Amazon connect while reducing call volumes by 15% and improved customer satisfaction.
• World fuel services uses AWS ML to provide predictive analytics for weather patterns and market predictions to decide whether to hedge or make investment in wind farms.