How I learned that Machine Learning was not enough
I have had the opportunity to work in different kinds of technology projects from Mobile, Web, Architecture Assessments, Internet of Things, and Machine Learning. I still remember working on my first Machine Learning project, I was super excited and it was way different than anything I have done in the past. Seeing the model learning and the different results we were getting was incredible for me.
After that, I was ready to keep on working on ML projects. I was ready to train, optimize, and deploy some models. But I found something very interesting in the process. There are a lot of mid steps between deploying where the client is and deploying your model. I was lucky enough to work on a project where a lot of these steps have been taken care of or ignored. I think in order to be a complete AI consultant you should be ready to work on any of these phases.
It all starts with the Strategy. Sometimes the client does not even know where to start. They want to do projects in the AI space, maybe they heard it in a conference or read it in the latest issue of a business magazine, but are not sure what they need, how to do it, or even why. In this phase, you need to walk the client through what is AI, the art of the possible, what they need, and what kind of benefits they might get. I am a firm believer that (almost) any company in any industry can benefit from AI. Although there are some foundational steps that can make the trip way more enjoyable.
After you’ve figured out Strategy, they might need to think about Data. Here you might need to ask some simple but powerful questions:
- Do they have data?
- Is it enough?
- Where is it stored?
- Can we get more data?
- Any privacy issues that we need to start thinking about?
Once these questions have been answered, we might need to actually do some things. You might need to move, transform, and load the data. We call this phase data integration, here SQL can be your friend. Although you might need more advanced skills like Spark or using a Data Lake. It will depend on the data you are using and how much do you need.
Depending on the client, you might need to figure out Infrastructure. A lot of clients are already on the cloud. Spinning a new database or using a PaaS service would be very easy. Others might be on the on-prem world, here creating a new database could imply buying a server. They might need to start thinking about buying GPUs (this might be a rare case, but could happen). Choosing the right cloud provider and moving data to the cloud can become complex projects but you can always try to start with something little.
Once you have figured out these phases you can start thinking about the part that you wanted to do in the first place: run, optimize, and deploy some cool Machine Learning Models. There is a lot of literature that covers this phase. What I strongly suggest is: don’t forget about the pre-requisites, they can make your ML Project way more complex than it should be.
If you like this story feel free to follow me on Twitter.
