This is how you land an ML job

I have worked for more than 10 companies in the last 9 years. Big and small. Startup and enterprise.

Sometimes doing ML
sometimes data science, and
sometimes doing data engineering.

I have been interviewed by senior devs. And I have interviewed junior devs, more than 20 times.

And over the years, I have learnt what makes a candidate stand out and get the job.

Let me share with you one secret.

Let’s start

The problem

One one side, there is you. You are looking for an ML job.

On the other side, a company that is looking for ML experts to help them solve their business problems.

And between you and this company there is one obstacle… THE 25-bullet-point job description.

 

So you feel a bit overwhelmed…

 

You feel you are not good enough (aka impostore syndrome) and your first instinct is to try to know just a little bit of everything either by

• doomscrolling tweets, posts and trying to read 20 blog posts a day…

 

• or starting your personal online-course-cerfiticate-collection-frenzy…

The problem is that this strategy does NOT work, because of 2 reasons:

1. Learning is not about following a happy-path-sequence-of-ready-made-exercises-with-solutions. Learning is about trying 20 wrong things, to finally find the right thing to do.
   So, no matter how many online course certificates you complete, you will still feel impostor syndrome when you start applying for jobs. And you know that.

2. Companies hiring data professionals could not care less about online course certificates. I have never been asked about what online courses have I completed, or what have I learnt from them. And I have never asked a candidate that question either.

   Companies are not abstract entities. Companies are people, like you and me, who need to find tech experts to solve their business problems. They don’t care about your online course certificates. What they want to see is if you can help them solve their business problems. In other words, they are looking for problem solvers.

   

    

So the question is:

The solution 

You need to swallow the red pill

 

The best candidates I have ever interviewed were those who had built something interesting in the past, and published it on their Github profile.

Seeing an actual project code and explanation is rare, and serves as a very strong signal that the guy knows what he/she is talking about.

Having a conversation with the candidate about it, is what reveals the thought process and problem solving skills of the candidate.

Now, the question is

Here is the trick 

You should not build 20 half-baked projects. Focus on ONE GOOD THING.

To find and build that project you need to play on 3 things:

• What you already know, for example, ML model training.

• What you don’t know but companies need, for example, ML model deployment to AWS.

• What real world problems companies need to solve, for example, improve customer retention.

For example

Let’s say that

• the last company you interview for used AWS to run their services, and

• you were rejected because you did not have enough AWS experience, doing things like ML model deployment.

You feel sad, but you also collected one precious piece of information.

“Companies out there need ML deployment skills on AWS” (which is by the way 100% true in most small-to-medium companies, and some big enterprises 😉)

Other examples of common tools you probably do not have experience with are:

• Apache Airflow,

• Kubernetes,

• Apache Kafka

• …

Now, you have 2 options here.

• You invest 6 months trying to get a cloud Certificate (like AWS or GCP). I personally think this won’t work for you, as preparing for these kind of certifications is more about learning how to pass their multiple-option tests, rather than getting hands-on experience doing actual work on the cloud, or

• Try to integrate the skill you missed (ML and deployments in AWS) into some project you’ve already done in the past. That is, you integrate (and learn) what you don’t know but need, into something you know and have already built in the past.

In this case, I would take one project I have built in the past, for example this taxi prediction service, and deploy its frontend UI not on Streamlit Cloud, but as a lambda function in AWS.

Moreover, if you know well what type of companies you want to work for, try to adjust the underlying problem.

For example, once you have built a system to predict taxi demand, you can adjust it to predict other time-series phenomena, like

• Stock market prices

• Electricity prices and demand

• Weather forecast

• etc, etc, etc (you get the point)

Changing the underlying problem requires more work, so you need to assess if it is worth the effort or not in your case.

In a nutshell

You build ONE project using:

• What you know (a project you have already built, so you don’t need to build things from scratch, but compound)

• What you didn’t know but needed (deploying to AWS)

• What companies need (time-series problems).

So you keep the design, but swap some pieces to target what companies look for.

You got the point.

Do you need help building your first real world ML project?

I teach 2 courses on Real World ML, one for batch-scoring systems and one for real-time ML systems.

The philosophy is always the same. I teach you how to design, build and deploy any ML system, so you can start building your own projects.

👉 If you have never built and ML system before, I recommend you start at The Real World ML Tutorial. It is a self-paced project in which I take you from what you know (aka training ML models in notebooks) to what has business value (aka an end-2-end ML system).

👉 If you already have some experience building ML systems, and want to take the next step and learn how to build Real Time ML Systems, I recommend you take a look at my live course Building a Real Time ML System. Together.

No BS.

Only the things I have learned at work, that companies need.

I hope this post helped you,

Talk to you soon,

Pau