The student data club is the new startup garage

It's 11pm on a Tuesday and a room in the engineering building is fuller than most lectures. Nobody is being graded. Six people are arguing about whether their model is leaking data, and they're right to.

This is what the best data work looks like right now, and it isn't happening where you'd expect. Not in a corporate lab, not in a lecture hall — but in student clubs that decided to act like startups instead of student clubs.

Clubs optimised for the wrong thing

The traditional student club is built around events and attendance. You show up, you sign in, you watch a talk, you leave. The metric is headcount. It's a model inherited from an era when access to information was the scarce resource.

But information isn't scarce anymore. What's scarce is the experience of building something real with other people who care — shipping, breaking, reviewing, and shipping again. That's the thing universities are worst at manufacturing, and it's the thing employers actually want.

We stopped measuring how many people showed up, and started measuring how many people shipped. Everything changed.

Six squads, not one committee

When we restructured the Data Science Club, we deleted the org chart and replaced it with six departments that behave like cross-functional squads — Data Insights, Core AI, Data Pipelines, Summits & Awards, Finance & Corporate Relations, and PR. Each one owns outcomes, not meetings.

A member can start in Data Insights making their first chart, and three months later be writing an ETL job in Data Pipelines that feeds a model in Core AI. The boundaries are real enough to give focus and porous enough to let people grow.

Here's the entire first version of one squad's feedback-scoring model — small enough to read in a meeting, real enough to ship:

# campus_pulse.py — score student feedback in real time
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import LogisticRegression
 
def train(feedback, labels):
    vec = TfidfVectorizer(ngram_range=(1, 2))
    X = vec.fit_transform(feedback)
    model = LogisticRegression(max_iter=1000)
    model.fit(X, labels)
    return vec, model

The whole model in twelve lines — a member shipped this in their first month.

What this means for how we learn

If the garage was where the last generation of companies was born, the student data club might be where this generation of builders is. Not because the tools are special — they're free and everywhere — but because the culture is. Treat students like builders, and they build.

That's the whole thesis. The rest is just sprints.