The $10K ML Competition Win That Changed How I Think About My Career

5 lessons about data science success that have nothing to do with code

I won $10,000 in USD Coin in a machine learning competition in Spring 2024, but never got around to writing about it, until last week. I finally wrote up the technical details of how I did it and Towards Data Science published it. It was fun, it almost felt like reliving the victory again!

But there’s more to the story beyond the technical details. In today’s post, I want to recap some of the bigger lessons that this experience taught me about data science careers.

Animation-style image of me coding (based on original image by author)

1. Domain Knowledge Beats Algorithmic Complexity

When faced with new problems, I like to draw on similar past experiences.

For example, I had never built a neural network for credit scoring before this competition that only accepted logistic regression or neural network models (for zero-knowledge proof verifications), but what I do have is extensive experience in finance, credit risk, and machine learning.

I also had never touched blockchain data before this competition, but I’ve worked day-in-day-out with financial transaction data.

Ultimately, understanding the business problem was most important and I treated this like any other credit scoring problem. My FinTech background was more valuable than any advanced degree, web3 knowledge, or fancy deep learning architecture.

2. I’ve Been Paying an “Imposter Syndrome Tax” for Years

Here’s something I hadn’t fully realized until I won:

I’ve been holding myself back because I don’t have the “right” credentials. No PhD. No computer science degree. No STEM background.

It isn’t without reason. I’ve had recruiters tell me point-blank that they only hire PhDs.

For years, this made me hesitant to:

• Apply for senior data science roles

• Enter competitions like this one

• Share my work publicly

Winning this competition was a reality check.

While I was worrying about not having the perfect academic background, I had something arguably more valuable: real-world experience solving actual business problems with proven results and impact.

I’ve passed up so many opportunities because I didn’t think I was qualified enough to even try. The “imposter syndrome tax” is expensive!

3. Start Before You Feel Ready

I almost didn’t enter the competition. My inner critic was extremely loud, telling me:

• “I don’t know Web3.”

• “I don’t know deep learning well enough.”

• “I’ve never done a data science competition before.”

But I decided to enter anyway for the experience. I made a commitment to myself to spend a weekend on it and submit whatever came out at the end of ~10 hours of work. This changed everything.

Looking back, I see this pattern everywhere in my career. The best opportunities came when I said yes before I felt fully prepared:

• Moving into FinTech (I did a masters in public policy)

• This competition (I knew nothing about blockchain data)

• My first data science role (I had little understanding of how production systems worked)

The confidence needed to overcome imposter syndrome comes from doing the work and achieving results to show yourself what you’re capable of. There’s no way to ever feel fully prepared before doing it.

4. External Validation Hits Differently

Winning this competition gave me something that a glowing performance review and annual raises never could:

Public proof that I have good data science chops.

It’s one thing for your manager to tell you you’re doing good work. But it’s another thing entirely to have a public win I can point to.

This shifted my thinking about ownership. As an employee, you build valuable models but can’t take them with you when you leave. You can’t charge royalty for every API call to your model. You don’t get ongoing credit for their success. This realization was part of what drove me to transition to consulting: I wanted to build something I could actually own.

5. Perfectionism is a Career Killer

I spent ~10 hours on this competition, but there was so much more I could’ve done: experimenting with different neural network architectures, hyperparameter tuning, feature engineering… the list goes on.

The perfectionist in me almost stopped myself from submitting the predictions and model, but I ended up following through on the promise I made to myself that I’d submit whatever came out at the end of the weekend.

This was an important lesson: I’d shipped an “MVP” of sorts, and it got me probably 80% of the way there. I could’ve spent additional days on it, but the extra 20% would have come with diminishing returns.

What This Means for Your Career

If any of this resonates, here’s what I wish someone had told me earlier:

Stop keeping your work private. Find ways to showcase your skills publicly.

Stop waiting for the perfect credentials. Start looking for projects or competitions in your area of expertise. Your domain knowledge is more valuable than you realize.

Stop giving in to imposter syndrome. You’ll never feel completely ready, just go for it and learn from the results. In retrospect, I’ve learned that the cost of not trying is often higher than the cost of failing or rejection.

The next time you see a role or competition in your domain area, don't think "I'm not qualified." Instead, think "I understand this problem better than most people." Give it a go and see where it takes you.

Want the Technical Details?

If you’re curious about the actual code, model development process, neural network architecture and techniques I used, I wrote a detailed technical walkthrough that got published on Towards Data Science earlier this week.

Read the full technical guide here →

It includes complete Python code, the feature selection strategy that cut features from 77 to 34, my neural network approach, and the threshold optimization technique that probably won me the competition.

Did any of this resonate with you? Hit reply or comment to let me know - I read every response.

---