Develop and Deploy Generative AI Applications on AWS with Eviden’s GenOps Framework - Part 2

In the previous article, we gave a general overview of the  Eviden’s GenOps Framework and the AWS Generative AI stack. In part 2, we will deep dive into the Design Stage. 

This stage involves 3 steps, and it consists of working with business stakeholders and technical partners to identify potential applications of Generative AI in business operations. This stage also involves obtaining some initial understanding about the project's scope, business requirements and workload, while also evaluating the tools and environment used by clients. 

Design decisions 

An important step in each generative AI project is defining the scope, including the specific generative AI use case and tasks that need to be addressed with the generative AI application. The funnel process is carried out through a series of workshops to guide the client during the use case thinking process and it includes the following main dimensions: the Positioning (quick wins vs disruptions), The Risk Assessment and the ROI Calculation (including build and run cost). 

There are also some other AWS recommended dimensions (detailed in Figure 2 bellow) that can be used to understand what the use case entails and think and guide the client about its design decisions. 

The first one is the criticality of the application that will be created, and this indicates the tolerance to mistakes.  For example, there is no problem if there is a mistake or losing some information when the use case concerns summarizing a book chapter or an article about a random topic. However, if the use case involves some sensible information such as summarizing credit card statements or fraud records, the tolerance to mistake will be much less. As a result, a low tolerance to mistakes obligates us to be more careful with the model selection which must be a higher accuracy model. 

The second dimension is the scale, and this indicates whether the generative AI application will have a targeted group or a broader audience.  

The third dimension is the task type, whether it is fixed or open-ended tasks.  A fixed type of task can be generally handled with smaller specialized models that are generally less costly. On the other hand, an open-ended type of task generally requires larger models that handle many tasks and need more knowledge. 

The fourth dimension is eloquence, which indicates the language fluency of the model that will be used.  

Another important aspect in defining the use case is assessing the long-term and short-term impacts of the use case. For example, clients can opt for an easy and quicker-to-market solution to improve developer productivity and start to realize benefits in a short time like embedding an AI coding assistant such as Amazon Q Developer into their workflow with a few fast steps. On the other hand, clients can opt for a robust and customized solution that would improve the customer’s experience like a chatbot which would take longer to implement and cost more if additional model training or fine-tuning is necessary.  

The figure 3 below illustrates the Use Case Long-term Vs Short-term impacts in the Eviden’s GenOps Framework. 

Quick wins use cases are consensual initiatives that lay the foundation for the innovation process. These are small-scale, low risk use cases that can be implemented quickly to generate tangible results and build momentum for larger innovative use cases. 

Development use cases help organizations have a reassuring mid to long-term vision that fosters the continuous progress of existing activities so they can effectively leverage their current capabilities and resources to drive sustainable growth. This approach enables organizations to maintain a competitive edge while ensuring the seamless evolution and advancement of their established activities. 

Disruptive use cases are a catalyst for rapid transformation and are characterized by their speed and compact nature, showcasing a strong desire to embrace change and serving as a powerful demonstration of the will to change. They often arise from unconventional thinking and a relentless pursuit of efficiency while also presenting novel solutions to longstanding problems. 

Transformation use cases represent radical and durable renewals for organizations seeking to unlock new territories and pave the way for unprecedented opportunities. These transformations use cases are rooted in a deep understanding of evolving customer needs, market dynamics, and societal shifts. By adapting and evolving their strategies, products, and services, organizations can maintain relevance and thrive in an ever-changing environment. 

Foundational Support Templates 

This step involves the development of a set of robust standardized templates that serve as the foundation for a successful Generative AI project implementation. These templates cover various aspects of the GenOps operational lifecycle, ensuring consistency and adherence to best practices across all teams involved. Those templates include for example: 

FinOps Templates

The FinOps templates are instrumental in driving financial and operational excellence. They cover key areas such as Operational Cost Savings and Revenue Uplift, enabling clients to measure the tangible benefits of their generative AI application. By providing a standardized approach to financial analysis, these templates empower decision-makers to make informed, data-driven choices that optimize the overall return on investment. 

POC/MVP Templates  

The POC/MVP (Proof of Concept/Minimum Viable Product) Templates serve as a roadmap for the development and deployment of generative AI applications. These templates outline the key components of the front-end, back-end, and architectural design, ensuring that teams can quickly and efficiently create prototypes and minimum viable products. This accelerates the iterative process of testing and refinement, ultimately leading to the successful launch of generative AI applications. 

Responsible AI Templates  

Responsible AI are crucial aspects of each Generative AI project implementation, as they ensure that the solutions adhere to regulatory requirements and uphold ethical principles. The Responsible AI Templates are developed based on new and existing requirements, incorporating a comprehensive risk assessment process. This approach helps clients establish a robust standard for managing data privacy, security, and the responsible use of generative AI. 

By leveraging these comprehensive templates, clients can unlock the true potential of the GenOps Framework, driving consistent, scalable, and efficient operations throughout the generative AI project lifecycle. These templates serve as a powerful tool for aligning teams, streamlining processes, and delivering exceptional results. 

GenOps Console 

The front-end of the application should be meticulously designed and prototyped to ensure a seamless and engaging user experience.  

The design process should be scoped based on the business use case and constraints, considering factors such as target audience, platform compatibility, and desired functionality.  

The front-end should be developed with a focus on visual appeal, accessibility, and ease of use, while also being tightly integrated with the back-end systems to ensure a cohesive and efficient overall application. We recommend using AWS Amplify or the Streamlit framework to develop the front-end. 

The following is an example list of features that can be included in the GenOps console: 

·       Access to Amazon SageMaker notebooks  

·       Collaborative Access (GitHub, GitLab, Jira/Scrum) 

·       Simplified GUI for the generative AI application POC 

·       Prototype of dashboard for monitoring/reporting 

·       User Authentication (SSO) 

Conclusion

In this blog article, we detailed the design stage of the GenOps 10 steps Framework leveraged by Eviden and its importance in each generative AI project.

The next blog article of this series will detail the Development Stage of the GenOps framework.