Machine Learning Autodidacts Need a Portfolio Equal to an $80K Degree. Here’s how:
Autodidacts focus on practical skills that ML professionals use every day. Once they have a core skillset and a micro portfolio, they apply to niche job opportunities that suit their no-degree background.
It’s all about momentum.
Self-learners often approach their careers in stages. First, they build a portfolio and get a foot in the door. Later, many make additional sprints to achieve career-specific goals. They either do them full-time, part-time, or through internal transitions and upskilling.
Here’s what those transitions can look like:
- Learning software engineering → A tech internship at a small company
- Building an ML portfolio → An entry-level ML role at a small company
- Part-time ML portfolio building → A junior ML role at a mid-sized company
- FAANG/GAFAM interview prep → ML engineer at a known tech company
This guide covers the first three steps and creates a foundation for the fourth step, FAANG/GAFAM jobs, but that’s out of scope for this guide. There are plenty of resources online covering interview preparation for large tech companies.
Old way: learn learn learn learn learn do
New way: learn do learn do learn do
- Wes Kao
The Recipe to Fail
I reckon effective self-learning mimics what ML professionals do on a daily basis. Yet, many self-learners do the opposite: they get stuck in long lists of online courses.
After gaining a handful of certificates, they run out of motivation.
They end up applying to the popular enter-level positions, but after not getting any interviews, confidence drops. They feel overwhelmed by **everything to learn** and have little hope of fixing it.
Here’s the thing.
There is little credential value in online certificates.
Online courses can provide structured learning resources but have a marginal impact on employment attractiveness. Most assessments have all answers available online, so there is little risk or consequence for cheating.
But the same is also true for most portfolios, many copy-paste or tweak existing projects. It’s hard to tell the difference between real and low-effort projects.
For companies, it’s too risky to advance candidates without evidence of being hireable.
After building a weak ML resume, many self-learners apply to known companies. High salaries and status seduce them. Viral Medium posts by autodidacts who land high-status jobs are often valuable. However, they lead to the wrong impression.
Early career high-status jobs without degrees are exceptions that are highly context-dependent.
Unfortunately, many self-learners don’t know how to find niche job opportunities that suit their no-degree background.
Instead, popular positions are flooded with applications. Many are complacent. With one-click applications and remote jobs, it’s easy to apply to hundreds of companies. Known tech companies have a hundred to several thousand applicants per position. In these resume lotteries, university graduates come out on top.
Many self-learners don’t land any interviews.
At this stage, it’s tough to get back on track. Self-learners spent their savings on living costs, they have little motivation left, and the pressure to create an income increases.
It’s hard to know what to do next.
Many ML experts recommend several years’ worth of online courses to reflect the depth of their education and career. Unfortunately, although it comes with good intent, it often sets unrealistic expectations on what it takes to enter the ML field.
At this point, self-learners don’t know if they should trust the academic camp that emphasizes calculus, algebra, statistics, and probability; the industry camp that argues for MLOps, pipelines, SQL, Git, and Kaggle; or the interview hacking camp that believes you should master LeetCode, cracking the coding interview, and memorizing the first part in Goodfellow’s Deep Learning book.
Self-learners often drop out at this stage.
Many never realize there are **far better ways** to land an ML job.
Many self-learners have a crucial misunderstanding: knowledge is not the same as evidence of being hireable.
This relationship was first studied by Solon and Hungerford in the late 80s.
For example, if a university student drops out a few weeks before graduating, employers don’t see it as 98% of a degree. Last-minute dropouts will have almost the same knowledge as graduates, but only be a fraction as hireable as someone who stayed a few more weeks.
While online courses are excellent learning resources, they seldom make a candidate more attractive to an employer. For self-learners, taking online courses is like someone who drops out just before graduation. They have the knowledge but not the hiring credibility.
If you don’t earn a traditional degree for your learning, you need to use your knowledge to compete for employment attractiveness.
Self-learners have to compete for either brand recognition, time, attention, or money.
When ML professionals validate your work, they are risking their time and reputation. If someone pays you for work, they are risking their money. If a conference publishes your paper, they are risking their brand.
When you compete for something scarce, you need trust, hard work, and talent. That’s what employers look for.
The hard part of the ML self-learning path is not how to gain knowledge but how to create industry credibility.
A portfolio is a collection of evidence that ML professionals and institutions have traded something scarce for your talent and hard work. That portfolio could include work experience, open-source contributions, or beating established benchmarks. They show that you have obtained something rare and valuable.