Shatter the Ice: Find Real Roommates

Our Story

Each one of us — at four different colleges — experienced the same anxiety during roommate season:

What if your roommate thinks your real interests are weird? What if you get stuck living with someone you don’t vibe with, for a whole year?

People hide parts of themselves to seem more “normal.” People often feel stuck with strangers they didn’t click with. And people feel the pressure to curate the “perfect” bio instead of the real them.

So we decided to fix the part of college that shouldn’t feel scary at all: finding the people you’ll call home.

Inspiration

Finding a roommate shouldn’t feel like a gamble.

Right now, students try their luck posting on Instagram pages, hoping someone reaches out. But those posts force you to filter out the quirky interests that make you you — because oversharing online can feel embarrassing or risky. Too many students end up living with people they barely know, leading to stress and conflict that could have been avoided.

Starting college is already overwhelming — new routines, new campus, new expectations. We built Shatter the Ice to make at least one thing easier. By letting users be honest without fear of judgment, the platform connects you with someone who actually understands you.

You deserve a roommate who gets your late-night study habits, gaming obsessions, or movie poster collection — without hiding who you are.

What it does

Shatter the Ice is a privacy-first roommate-matching platform using semantic AI to match students based on:

Real, private interests
Living style compatibility
Hard filters like gender preference and pet allergies

There are two main ways to find your match:

1. Instagram Profile Links Add your private Shatter the Ice link into your bio. Whoever clicks sees instantly whether you’re compatible — no awkward guessing.

2. Connect Recommendations A ranked list of your top potential roommates based on AI matching. No posting. No waiting. Real matches come to you.

When a student chooses to “shatter the ice,” the platform reveals:

Shared interests
Related interests (semantic similarity)
A combined match score
AI-generated conversation starters

Above 50% compatibility, messaging unlocks — signaling a match worth pursuing.

How we built it

Backend

Node.js + Express
MongoDB + Mongoose
JWT for secure auth
Claude Haiku 4.5 for similarity + AI prompts
Match caching (Mongo + in-memory)

Frontend

React + Vite + Tailwind CSS
Three.js ice-shattering animation
Responsive glass-morphism UI
Socket.IO for real-time messaging with offline persistence

Deployment

Client and API deployed separately on Heroku
Secure environment variable configuration

The result: fast, scalable, privacy-respecting performance.

The Matching Algorithm

Two core scores determine overall compatibility:

1) Interest Score (0–100%)

We embed each interest using Claude and:

Compute a similarity matrix between all pairs of interests
Take each interest’s best match, which prevents cheating with duplicates
Average top matches for a base score

Bonus logic rewards real alignment:

+5 for each extremely strong match (90+)
+5 if ≥3 strong matches (70+)
+5–7 based on total aligned interests

Quality and depth matter more than list length; the score is capped to prevent inflation.

2) Lifestyle Score (0–100)

Seven weighted factors measuring real-world roommate fit:

Sleep schedule
Cleanliness
Social energy
Guest frequency
Bed/wake time differences
Pets

These are normalized to an overall score of 100.

This avoids great-interest matches living terribly together.

3) Hard Filters First

These are enforced before any scoring:

Gender preference (bidirectional)
Pet allergy compatibility
Only show matches within the same campus ecosystem

Unviable matches never appear in the first place.

4) Combined Score (Displayed)

The combined score is weighted 60% from interests and 40% from lifestyle. We chose to do this because we believe that while interest builds friendship, lifestyle prevents conflict.

Reveal thresholds: