Hackathons and Kaggle for freshers: what they prove, how to show it

What an interviewer infers from each

The key interview risk is the follow-up. "You were in the top 10%. What would you do if the test distribution shifted?" If your answer is only about the leaderboard, the line helps less than it looks.

Resume lines, weak and strong

Kaggle, weak: "Kaggle Competitions Expert. Participated in 8 competitions."

Kaggle, strong: "Tabular regression competition (house prices, public dataset): built a gradient-boosted model with target-encoded location features; 5-fold CV RMSE matched the private leaderboard within 2%, placing in the top 15% of teams. Notebook with validation write-up linked."

Why it works: it names the task, the technique, the validation method, the fact that CV tracked the hidden test score (which shows you didn't overfit the public board), and the rank last.

Hackathon, weak: "Participated in college hackathon, 2026."

Hackathon, strong: "36-hour college hackathon, team of four: prototype that flags duplicate complaints in a city grievance CSV using TF-IDF similarity. I wrote the cleaning pipeline and the precision check on 200 hand-labelled pairs. Finalist; demo video and repository linked."

Never claim a teammate's work. Interviewers often ask, "Which part did you write?" and then ask you to explain it.

Check that your validation was honest

The strongest Kaggle story is that your own validation predicted the hidden score. You can show this in a small table. Here is how to compute the gap from your own submission log; the numbers are hypothetical.

# Hypothetical submission log: your cross-validation score vs the private leaderboard score.
# Lower RMSE is better.
subs = [
    ("baseline linear",       0.1620, 0.1655),
    ("gbm default",           0.1310, 0.1342),
    ("gbm + target encoding", 0.1215, 0.1238),
    ("gbm tuned on public LB", 0.1180, 0.1291),
]

for name, cv, private in subs:
    gap = (private - cv) / cv * 100
    flag = "  <- CV no longer trusted" if gap > 5 else ""
    print(f"{name:24s} CV {cv:.4f}  private {private:.4f}  gap {gap:5.1f}%{flag}")

The last row is the classic story: tuning against the public leaderboard improved the CV number but the private score got worse, a gap of about 9%. Being able to explain that row, and why you'd have picked the third submission, is worth more in an interview than the rank itself.

Choosing what to spend time on

• If you have no projects at all, do one end-to-end project first (question, data, cleaning, analysis, write-up), then a competition. The final-year project rubric helps you pick one.
• If you're aiming at ML roles, two competitions where you can explain validation choices are useful. Ten shallow entries are not.
• If you're aiming at analyst roles, hackathons with business-flavoured problems and a presentation round usually map better than leaderboard modelling.
• Prefer completed competitions for learning: winning solutions and discussion threads show you what you missed, which is the fastest feedback you'll get for free.

Presenting it in the interview

Prepare a 60-second version for each item: the problem, your approach, how you validated it, one mistake and what you learned, and your personal part. Then prepare for these follow-ups:

1. 1Why that metric, and what would the business prefer instead?
2. 2What did the top solutions do that you didn't?
3. 3How would this run on new data every week?
4. 4What would you cut if you had half the time?

If you can answer question 3, you're already thinking beyond competitions. The Data Science programme spends much of its six months on exactly that gap, from framing a problem to deploying and monitoring a model, with mentor review of each project.