Infrastructure
1. Infrastructure
With an understanding of the problem, goal, and data - let's determine the appropriate technical design and requirements for the infrastructure.2. To build or to buy?
First, it's important to understand if the goal necessitates a completely custom AI solution or leveraging an existing product or service. A custom solution will allow a business to have more control over the entire system and make it hyper-specific to their particular use cases. It will require more strategy for the design, staffing, development, and maintenance. There may be longer timelines and higher costs as well. Finally, the AI solution is not guaranteed to be best-in-class, as there's a lot of great companies focused on these types of technologies. Buying an existing product or service will generally be usable out of the box. It requires less specialized skills, such as AI and machine learning, among the project team but experts should still be involved. Depending on the scenario, time may need to be spent customizing the AI with data. Likewise, more time may be needed for integration into existing systems. A hybrid solution using both custom and purchased components could also be built.3. Choosing an environment
The other question to determine is in what type of environment to build the solution - in the cloud or on-premise. Again, the choice here will impact the time, energy, cost, and skill set required. It will also determine how the next components are designed. With that said, modern AI solutions are typically built in the cloud.4. Technical requirements
AI solutions differ in the kind of hardware that is required. Typically, it will need to be able to handle higher volumes of data, more complexity, as well as typical software processes.5. Technical requirements
The average AI POC will include the following components - data infrastructure,6. Technical requirements
an AI development environment,7. Technical requirements
and a deployment infrastructure.8. Data infrastructure
Data infrastructure includes components such as storage9. Data infrastructure
and data pipelines. If we're using our own data, we may already have the foundations in place. Even so, the storage system may need to be updated for increased storage capacity and faster data retrieval.10. Data infrastructure
The data pipelines allow the AI model to access the data in order to learn the patterns. It's important to nail down this data infrastructure because this will determine how fast the AI in the solution can be developed.11. AI development environment
This brings us to the AI development environment which is the core of the solution. This will be where the AI model is developed to learn patterns. Here are some factors to think about.12. AI development environment
Compute, or how much power is required for training the AI. The amount of data and the complexity of most AI algorithms, requires specialized hardware, such as graphic processing unit chips, or GPUs.13. AI development environment
The algorithm itself. If we're building a custom solution, this is where a machine learning or AI expert will be vital. They will help determine the most appropriate modeling techniques to meet the use case objectives.14. Deployment infrastructure
The deployment infrastructure is where the AI solution will be hosted, interacted with, monitored, secured, and more. Scaling from a POC, it will also need a system for Machine Learning and Development Operations. Questions to consider are the following. What is the end-user interface? Does it need to be built or will the AI solution be integrated into an existing system? Does the AI solution need to be real-time? Who should be able to access the AI solution? Answers to these questions will help determine the type of compute power, networking, security, and interface required.15. Let's practice!
It will be important to bring in the necessary experts to help build out the detailed specifications. We will talk about these roles in the next lesson.Create Your Free Account
or
By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.