Machine Learning Operations on AWS

AWS will advise and assist Customer with the following activities:

• Provide an overview of MLOps and ML industrialization
• Review Customer’s current state ML industrialization strategy, and provide a documented assessment including Customer’s operating model, limitations, and requirements
• Define Customer’s end state ML industrialization strategy
• Provide Customer with hands-on workshop with an ML industrialization platform leveraging AWS Services including Amazon SageMaker (including SageMaker Pipelines), AWS Glue, AWS Lambda, AWS CodeCommit, AWS CodePipeline and AWS Step Functions. During this workshop, AWS will provide source code of a ML use case example using public data, whch can be used as a baseline for future ML solutions by Customer.
• Provide Customer with a high-level roadmap for the end state ML industrialization strategy, including key milestones and actionable items for Customer
• Provide Customer with a high-level technical roadmap for Customer to implement the end state ML industrialization platform

AWS will provide the following deliverables to the Customer during the engagement (“Deliverables”):

• Documented assessment of the current-state ML industrialization strategy, including Customer’s operating model, limitations and requirements
• Documented high-level roadmap for the end-state ML industrialization strategy, including key milestones and actionable items for Customer
• Documented high-level technical roadmap for Customer to implement the end state ML industrialization platform
• Source code of an ML use case example using public data

AWS will advise and assist Customer with the following activities, in Customer’s non-production environment:

• Define Customer’s end-state AI/ML design and AI/MLOps processes, based on Customer’s business and technical requirements
• Create a high-level technical implementation roadmap for the AI/ML use case
• Implement Data Innovation Workbench and Configure the AI Foundation Solution Kit (DIW AI Foundations SK)
• Deploy pre-packaged infrastructure needed to leverage selected AI/ML Models for the Solution

Utilize Data Innovation Workbench and AI Foundation Solution Kit for available Infrastructure as Code (IaC) and CI/CD templates to perform the following:

• Design and build the target architecture for Customer’s existing ML use case in a non-production environment Assist Customer’s cloud infrastructure team to:
  • Set up AI/MLOps and data science accounts using a multi-account structure in a non-production environment
  • Set up networking architecture and identity and access management (IAM) with associated IAM roles and policies
• Advise on code repository structure for the in-scope AI/ML use case as identified by Customer, based on AWS coding general best practices
• Define workflow orchestration, such as AI/ML pipelines for training and inference for the AI/ML use case
• Configure DIW AI Foundations SK to test, and deploy continuous integration and continuous deployment (CI/CD) pipelines for the AI/ML use case in a non-production environment
• Develop as custom Infrastructure as code (IaC) where appropriate and not available through DIW
• Create a storage mechanism for generated AI/ML models and metadata
• Deploying, in a non-production environment, AI/ML use-case-related infrastructure to multiple AWS accounts using CI/CD pipelines with development, testing, and pre-production stages
• Implement model monitoring functionality to detect AI/ML model and data drift

AWS will provide the following deliverables to the Customer during the engagement (“Deliverables”):

• Documented architecture for model development, deployment pipeline, and inference pipeline
• Data Innovation Workbench and AI Foundation Solution Kit
• Configurations for setting MLOps in non-prod environments using Data Innovation Workbench
• Any custom code or CI/CD pipelines developed by ProServe outside of the DIW AI Foundation Solution Kit will be provided to the customer.
• Customer will be solely responsible for all activities in production environments.

Customer will:

• Provide anonymized sample data for use in non-production Amazon Simple Storage Service (Amazon S3) bucket(s)  
  • Data must be in JSON, CSV, Parquet, Avro, TXT (text data), JPEG or PNG (image data), MPEG-4 or MOV (video data), 16kHz or 8kHz audio streams (audio data)  
  • There must be one (1) week or less of data processing work involved based on the staffed AWS’s data scientist’s estimate. If the estimate is longer than one week, Customer is required to apply necessary data processing in advance of the engagement  
  • Data provided must contain sufficient relevant data to inform meaningful predictions in terms of both rich features and historical data  
  • Customer will provide a description of data sources and attributes
• Provide support and approval from its business stakeholders, application owners, information technology (IT) infrastructure team, and security, as well as any necessary exceptions to standard processes to execute on the Project
• Ensure active participation of Customer subject matter experts and full-time staffing of Customer employees, contractors, and third-parties as necessary to successfully execute on the Project
• Establish and make available a communication plan and escalation path to quickly resolve risks and issues in connection with this Project
• Provide access to equipment, source code, data, and facilities necessary to successfully execute on the Project
• Make available on a timely basis any Customer staff who will support AWS in the execution of the tasks described in this SOW
• Establish a dedicated team to work with the consultants provided by AWS on the Project contemplated by this SOW
• Provide feedback to AWS in a timely manner to allow the Project to proceed in accordance with any agreed timelines. AWS’s compliance with the estimated timelines stated under this SOW is dependent on Customer’s timely co-operation under this SOW
• Prioritize the Project contemplated by this SOW, and provide the resources required to produce the deliverables and achieve the desired outcomes under this SOW
• Commit to changing or creating policy exceptions to achieve the Project outcomes as required
• Implement any required security firewall configuration changes and obtain security policy approval to support the Project as required
• Conduct an hour of weekly meetings with technical and partner teams
• Ensure that a suitable AWS data ingestion, data analytics and data science environment is available prior to commencement of project
• Provide subject matter experts to support AWS on data and ML use case algorithms/code understanding and acquisition activities
• Ensure availability of the necessary AWS accounts prior to the start of the Project
• AWS will support Customer in obtaining approval of the proposed architecture with Customer’s security and compliance teams for production usage. The sole responsibility of having the proposed architecture vetted and approved with Customer’s internal security and compliance teams lies with Customer
• Configure Amazon services used as appropriate for Customer’s workloads, and take steps within its control to maintain appropriate security, protection, backup and routine archiving of Customer content
• Agree with AWS how the ML use case will be deployed
• Ensure all required data is available in Amazon Simple Storage Service (Amazon S3), and the ML use case code is accessible to AWS during the term of this SOW
• Will be solely responsible for any required initial data gathering and transformations, including extract, transform and load (ETL) workloads
• Will agree to and support the creation of public referenceable artifacts such as a reference slide, a case study, a blog post, or a press article