So you have built your machine learning model, now what? How do you make it usable by your customers?
You’ll have to deploy it on production and build a web application so customers can interact with it. Use the deployment and web application calculator below to estimate your cost.
Good question. There are 3 reasons why we can do this:
We are currently building a platform that automates this entire workflow, and while we aren’t quite there yet, we’ve automated a few parts of it. This helps us deliver faster and reduce cost.
We have deployed many machine learning models on production and have learned a lot about best practices to make this process faster and common mistakes and pitfalls.
For example, we know that most people use standard frameworks and libraries such as Tensorflow, Pytorch or Sctikit-learn (sklearn) to build machine learning models. We also know that most people deploy their machine learning models on cloud service providers such as AWS, GCP or Microsoft Azure. And finally, most people use a MERN or MEAN stack to build a modern web application to demo their models. This experience helps us standardize and create reusable components.
We’ll be building most of the components you need for your web application anyway so we can offer them to you at a fraction of the cost while we get customer feedback to improve our platform.
That’s always an option. Here’s some information on how to find the right people, rough timelines and budget.
Machine learning model deployment requires expertise in multiple areas such as cloud infrastructure, ML engineering and full-stack web development.
An area to watch out for is that often ML models are deployed by either data scientists who are good at converting data into useful models but not necessarily trained for software engineering, or by web developers who are great at building responsive applications but don’t understand the importance of data collection and processing in machine learning. Therefore, it’s important to know what skillsets are required to get this done.
Keep in mind that machine learning model deployment is only half of the story. Depending on what you are trying to accomplish this may be enough. For example, if you just need to show a proof of concept to a potential customer or need an MVP to raise capital.
However, to build a truly data-driven solution, you would have to consider dataflows, pipelines and build the infrastructure around it. Furthermore, to build an AI driven application that you can use to serve your clients, you might need to hire a cloud architect to set up cloud infrastructure, a data engineer and machine learning engineer to set up data pipelines and to deploy the model, and a full stack web developer to build an application and connect it to your model via APIs.
In terms of budget, each specialist might cost you anywhere between $60K-$100K. To find them, you’d spend around 1-3 months. To build your complete application, you’d add another 2-4 months for a total of 5-8 months.