I hired ML engineers with no ML experience. Here's what substituted for it
The fresher's trap: no ML job without ML experience, no ML experience without a job. I have hired people straight out of that loop. Not on potential, and not on certificates. On three specific things that stood in for experience, and that anyone can produce in a few weekends without a job.
The fresher’s trap is real and it is circular. You cannot get the ML job without ML experience, and you cannot get the experience without the job. Every student I talk to is stuck somewhere in that loop, usually adding another certificate to the pile and hoping it breaks.
It does not break with certificates. I have hired people with no formal ML background and no industry experience, and I did it because they handed me something that stood in for experience better than a job title would have. Not potential, which is unfalsifiable, and not a longer list of courses, which is just proof you can enrol. Three specific things, all of which you can produce in a few weekends without anyone’s permission.
One shipped project you can defend
Not a portfolio of ten half-things. One project that reached a real user and that you can talk about honestly, including what is broken in it. I have written separately about what actually makes a project stand out to the person screening you, so I will not repeat it here, except to say the bar is “someone who is not you used it,” not “it got a good score.” Experience is just a record of having built things that mattered. A shipped project is the smallest possible version of that record, and it counts.
One piece of writing that shows you can think
This is the one candidates skip and it is the one that moves me most. Write up a decision you made and reasoned through, publicly. Why you chose one approach over another, what went wrong, what you would do differently. A blog post explaining a tradeoff is proof of the thing a job interview is trying to find out and usually cannot: whether there is structured thinking behind the code, or whether you were pattern-matching. Anyone can list “PyTorch” on a resume. Almost nobody can point to a page where they reasoned in public, which is exactly why it works.
If that sounds circular, notice what you are reading. This post is the artifact I am describing. A public explanation of how you think is the credential. You do not need a company to grant you permission to produce one.
One real problem, not a course assignment
The difference between “I did the capstone from the course” and “I built a thing because a problem annoyed me” is visible from across the room. The first is compliance. The second is initiative, and initiative is the trait I am actually hiring for, because I cannot teach it and the job runs on it. Pick a problem in your own life, your college, your side hobby, and solve a slice of it with ML. The problem being small and real beats the problem being impressive and borrowed.
What does not substitute
More certificates. More courses. A longer skills section. These feel like progress because they are measurable and because completing them gives you a certificate-shaped hit of accomplishment. But they answer a question I did not ask. I am not asking whether you studied ML. I am asking whether you can do the work, and the only evidence for that is work. Shipped, explained, and yours.
The loop feels closed from inside it. It is not. You break it by producing the evidence a job would have given you, before the job. Most people never try because it is harder than enrolling in one more course. That is precisely why it works when you do.