Is Google Vertex AI right for your project?

Are you in the process of deciding whether to implement Google Vertex AI? Our Senior AI Developer, Nic von Ahsen, has put together this guide to choosing whether Vertex AI is the right fit for your project.

Google Vertex AI describes itself as ‘a unified UI for the entire ML workflow’; in simplistic terms, it’s a centralised platform for ML operations. It’s a great tool for companies who don’t have an AI developer specifically, are looking to reduce the hassle of employee turnover, or want to reduce the number of people needed for a project.

Vertex AI has a lot of advantages: centralisation, accessibility, and reducing the amount of operational work needed.

Vertex AI as a platform promotes visibility and transparency, making it a great tool for a developer who’s new to AI and wants to see what their options are within ML operations. Its visibility and centralisation also makes it much easier to hand over your work to another person. If you need to swap developers, Vertex AI makes it easy to pick up where the last person left off because of a. how visible and streamlined the platform’s options are, and b. because Google requires you to work their way, meaning there will be less difference in two developers’ workflow and code.

Equally, it’s great for teams that want to hand over the maintenance once the initial project has been completed. Because of how accessible Vertex AI is to developers who are new to AI specifically, it allows for the ML developer to hand over the project to a company that doesn’t have a full time AI developer on staff. What once required a team of ten, now becomes a team of one or two.

Another big payoff for Vertex AI is that ultimately, if you play by Vertex AI’s rules and systems, a significant amount of the operational work is done for you. However, adjusting to Vertex AI’s way of operating can be a disadvantage in some cases depending on the developer.

Vertex AI can be a hassle to transfer pre-existing code and workflows into.

Because you have to play by Vertex AI’s rules, it means you can become locked into Vertex’s way of doing things. If you’re bringing in old code or code you’ve developed before transferring to Vertex AI, you’ll have to redo it to fit in with Vertex’s. There may be difficulties working on certain tasks or libraries that don’t play nice with how Vertex AI works. This also means that if you build something with Vertex and want to transfer it outside the platform, it may not work either.

So who should be using it?

There are three key groups of people who would benefit the most from using Vertex AI:

1. Developers who are new to AI and don’t know what their options are within ML operations yet.

Because of how centralised and clearly laid out Vertex AI is, it’s great for people who want to play around and learn what options are available within ML operations. A lot of your work can be done through a visual user interface and using AutoML, meaning you can avoid writing much code. The barrier to entry is reduced in that you no longer need a team to implement separate infrastructure; with Vertex AI, everything is in one place and accessible to developers to AI newbies.

2. Small to medium enterprises who want ML architecture, but don’t have a team of people for it.

Normally you would need an Operations developer to set up servers, a backend developer to set up APIs, a data-scientist or ML developer, and a method to facilitate communication between them. All of these are different skill sets, and these developers would normally be working in separate, decentralised platforms with little visibility within the different stages and tests. If you use Vertex AI right, you only need one AI developer to implement your project.

There’s also no worry around a developer leaving and having to transfer their work - with Vertex AI, it’s all in one place, and set to one standard.

3. AI developers, data scientists, or researchers, who have either trained a model or have datasets and want to deploy them in a scalable, maintainable way, but don’t know their way around Cloud and operational tasks.

Vertex AI aims to consolidate and simplify all the operational tasks related to deploying, versioning, and maintaining models and datasets. Data scientists and researchers who specialise in developing models often won’t have the operational experience to make the most of cloud providers’ offerings and operations best practices.

Vertex brings all of these tools into a single place with documentation and a nice user interface. As such, it’s a great way to start aligning your model with best practices when you’re looking to move out of the lab and onto the web.

Why Vertex AI over other platforms?

It’s less a matter of Google Vertex AI vs. Stagemaker or Studio XX; none stand out more than the others. The question should be more focused on why someone should be using one of them as opposed to not. Vertex AI - as well as Stagemaker and Studio XX - should be seen more as a useful tool to developers within ML operations.

The benefits of having ML operations in a centralised location ultimately outweighs the downsides - though, this depends entirely on how skilled someone is within AI. A centralised ML operation works best for those new to AI,  or people who already have models and datasets; a large team of trained developers would benefit more from their current system.

For companies already working within the Cloud, Vertex AI is an option for improving and simplifying operational processes, or to simply get into ML in the first place. The appeal of Vertex AI is in its lowered barrier to entry; with Vertex AI, having an ML operations developer or full-stack engineer is optional.