Generative AI on AWS

• Provide an overview of generative artificial intelligence (“GenAI”) on AWS, expected benefits and typical use cases
• Discover up to three mutually agreed upon GenAI use cases for Customer and document Customer’s business and functional requirements, pain points, gaps, and Customer’s desired business outcomes
• Review and provide recommendations, based on AWS general best practices on Customer's existing technologies, data and machine learning environment, and operational requirements concerning the mutually agreed upon GenAI use cases
• Review data dependencies for the mutually agreed upon GenAI use cases
• Assist Customer in performing an assessment of the Customer’s capabilities to implement the mutually agreed upon GenAI use cases.
• Assist Customer in developing a business case for the mutually agreed upon GenAI use cases with potential business outcomes and success metrics.
• Assist Customer in creating an implementation plan for the mutually agreed upon GenAI use cases.

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

• Documents
• Documented proposals and next steps to implement/operationalize up to three (3) mutually agreed upon use cases
• Up to three (3) documented, prioritized use cases, including business case with potential business outcomes and success metrics
• Documented recommendations for technology capabilities and processes required to support the mutually agreed upon use cases
• Documented business case with potential business outcomes and success metrics.
• Documented high-level roadmap to implement selected mutually agreed upon use case(s)
• Documented risks and challenges for selected mutually agreed upon use case(s)

• This engagement does not include any form of implementation or coding work

The Customer will be responsible for the following:

• The Customer is required to designate the following individuals to participate
  • Sponsor who owns the overall business or operational target and end-vision of the target state
  • Business Decision Makers (BDM)  (Product Managers) who have an in-depth understanding of the business and operational processes relevant to the use case discussed.
  •  Technical Decision Makers (TDM) (Data Scientist/Architects/Engineer) who has a comprehensive grasp of the data sources and infrastructure that are pertinent to the use case being discussed
• Providing a description of data sources and attributes  
• Providing integration requirements with any existing legacy applications
• Providing legal, compliance, and security requirements pertinent to the use cases being discussed

More specifically, AWS will assist Customer with the following activities:

• Review Customer’s existing data, machine learning environment, and future state infrastructure requirements, business and functional requirements, operational requirements, pain points, gaps, and Customer’s desired business outcomes
• Define and design the Solution which may include the following:
  • Customize one or more foundation models (FMs) on Customer’s data through fine-tuning,
  • Prompt engineering
  • Retrieval augmented generation (RAG)
• Define a high-level architecture of AWS Services in relation to Solution
• Explore and evaluate potential FMs in relation to the Solution and Customer data
• Develop repeatable training scripts for adapting the FM for the Solution
• Evaluate the model’s performance as it relates to the Solution and document the results
• Conduct knowledge transfer of the Solution design to Customer
• Create mutually agreed upon pipelines or infrastructure needed to leverage the GenAI model for the Solution.
• Advise and assist Customer with developing, testing, and deploying continuous integration and continuous deployment (CI/CD) pipelines for the Solution in a non-production environment
• Advise Customer on defining operating model for the Solution

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

• Documented Source code to prepare the Customer’s data, fine tune the FM if needed, evaluate the solution’s performance, and deploy the Solution to the Customer’s non-production environment.
• Documentation of the model’s performance and the steps that AWS took to prepare the data, select the FM, and train (if necessary) the FM.
• Basic automation to deploy the GenAI model and related infrastructure in the Customer’s AWS account.
• Documented recommendations, based on AWS general best practices for next steps

• Customer has undergone the “Generative AI on AWS Discovery” offering
• Customer has a non-production Sandbox AWS account & landing zone built
• Customer has completed all necessary data preparation, labelling, and structuring in the required format for model customization
• Customer has SMEs as well as business and technical owners available to review the results
• Customer has the services enabled in their AWS environment to facilitate AWS Professional Services to deliver the items described in this statement of work. 

• Customer will make relevant data accessible to AWS team
• Customer is responsible for the safe and ethical use of the Solution including the implementation of additional checks or acceptance testing on the model output
• For all ML models used during this engagement, Customer will review and abide by all relevant model licensing or usage terms
• Customer will be responsible for Customer’s staff who will support AWS in the execution of the tasks described in this SOW

• Customer has requested that AWS provide to Customer a proof of concept or solution under this SOW (“Solution”) that may include use of or access to machine learning models (“ML Models”). ML Models may be based on a third-party model such as those available in Amazon SageMaker, Amazon Bedrock, or elsewhere (e.g., Hugging Face). For clarity, third-party ML Models constitute Third-Party Content under the Agreement, which may be subject to separate license terms, usage restrictions, and other terms and conditions.
• For clarity, as between AWS and Customer, output generated by an ML Model constitutes Customer Content.
• If Customer provides Customer Content to AWS for further training, fine tuning, prompt engineering, retrieval augmented generation, or similar activities with respect to an ML Model, Customer confirms that it has all necessary rights in such Customer Content for AWS to perform the AWS Professional Services under this SOW
• Without any limitation to the Agreement, AWS will not receive any right, title, or interest in or to any incremental improvements, enhancements, or other modifications created in the fine-tuned models as a result of having been fine-tuned under this SOW (such incremental improvements, enhancements, or other modifications, the “Incremental Fine-Tunings”), to the extent that any protectable intellectual property rights in the Incremental Fine-Tunings are created as a result of the fine tuning under this SOW. The Incremental Fine-Tunings exclude the FM model from which the Incremental Fine-Tunings were created.
• Customer acknowledges that the Solution is not designed or intended to (a) make automated decisions that could have a consequential impact on an individual’s legal or financial position, life or employment opportunities, human rights, or result in physical or psychological injury to an individual, (b) support any use in which an interruption, defect, error, or other failure could result in the death or serious bodily injury of any individual or in physical or environmental damage, or (c) meet Customer’s regulatory, legal, or other obligations. Customer is solely responsible for conducting additional testing, assessments, and implementing use case-specific safeguards, as needed.