Enterprise AI High Level Design for Enterprise AI Solutions
In AI-naive organizations, AI point solutions don’t scale to the Enterprise. Thereby, it’s important to incorporate Enterprise AI HLDs which maps to the ML reference architecture and AI/ML platform.
When starting out with Enterprise Data Science at your organization, you may find that your company has already hired data scientists to build AI point solutions in the various departments, for a specific use case. However, when they try to scale their point solution to the Enterprise, the individual AI project teams don’t have the proper solutions architecture for their AI project to deploy in production.
Your organization may also have a machine learning reference architecture (MLRA), as your IT architects have anticipated the target state where the data scientists would want to eventually scale their AI solutions at Enterprise scale. The MLRA then provides the blueprints and IT services needed for the various architectural patterns that encapsulates most of the AI/ML solutions that your organization may want to operationalize.
Additionally, your AI/ML infrastructure team may have already anticipated the need for a centralized, Enterprise AI/ML platform to deploy AI/ML models in production, and built a Candidate Architecture Recommendation (CAR) with a High Level Application Interface View (HLAIV) and Technology Architecture (TA) view. This CAR is the blueprint for the platform.
So you have your enterprise blueprints with the MLRA and platform CAR with HLAIV/TA view. But then you go to the project team level, and find that the AI project solutions architecture can’t scale and doesn’t align to the MLRA and platform CAR.
A solution to this problem is for the infrastructure/MLOps team to architect an Enterprise AI High Level Design (HLD) which aligns to the MLRA and platform CAR at the Enterprise, while also incorporating the requirements of the AI solution at the project level. This Enterprise AI HLD then serves to help AI project solutions architecture to scale appropriately in line with the MLRA and platform CAR.
The problem with the MLRA is that it may be too detailed for the project level. And the problem with the AI project solutions architecture is that it may be too high level, where they don’t describe all of the workflows and pipelines needed for the end to end data science lifecycle. I find that an Enterprise AI HLD hits it just right, and the appropriate granularity.
So what does an Enterprise AI HLD look like? It has the main components of a complete AI system, which includes the following 3 pipelines (3 pipeline design):
Feature pipeline
ML Training pipeline
Inference pipeline
And these pipelines are accompanied by a Feature Store and Model Registry. These components forms the basis for this Enterprise AI HLD, and you will have different HLDs depending the the use case:
Tabular
Text
Computer Vision
And depending on the data refresh rate:
Batch
Real-time (or near)
So if you have a tabular use case for the AI project, and the predictions are consumed by the application every morning, then your Enterprise AI HLD would be for this tabular use case, batch. So your HLD will change according to your project use case category and data refresh rate.
So when your infrastructure/MLOps team provides this Enterprise AI HLD for your AI project, it improves your own AI project solutions architecture and allows it to scale at the Enterprise, while also complying with the MLRA and platform CAR.
If you're looking for support from me, here are a few options:
Enterprise Data Science Consultancy: With my consult team comprised of a Senior Data Scientist, Senior ML Engineer, Senior Data Engineer, and Senior Cloud Engineer, we will help you architect and build your Enterprise Data Science platform, and transfer knowledge to your IT team to maintain and optimize it. We will also overlay an MLOps framework to manage the AI solutions you build on this platform. If you don’t have an MLOps team, we will help you build one. Please get in touch about this consultancy here
Coaching and Mentorship: I offer coaching and mentorship; book a coaching session here