Zepic

Zepic maps campuses into a skill graph, matching students with “weak ties” who spark new ideas—plus AI insights, paths, and a 7-day plan to turn intros into action.

Comment

Inspiration

Most students network inside the same bubbles: same major, same classes, same clubs. But research—like Mark Granovetter’s “Strength of Weak Ties” and large-scale studies from LinkedIn—shows that weak ties often create the most valuable opportunities.

On campus, those connections are invisible. The designer who could unlock a research project, the ML student a bio major needs, the one person who bridges two disconnected communities—students rarely find them.

We built Zepic to make those connections visible and actionable.

What it does

Zepic turns a campus into a skill-interest graph and routes students to high-value “weak ties”—people who are different enough to unlock new ideas, but connected enough to make a first conversation work.

Students:

  • Log in securely with Auth0
  • Create a 60-second profile (skills, interests, goals)
  • Receive AI-ranked matches based on:
    • Brokerage (bridging different clusters)
    • Complementarity (non-redundant skills)
    • Serendipity (useful, surprising overlap)

Each match includes:

  • A graph path explaining the connection
  • Evidence (shared signals)
  • A 7-day activation plan
  • A custom icebreaker

This turns networking from vague intent into immediate action.

How we built it

Frontend

  • Next.js (App Router)
  • React
  • Tailwind CSS
  • Component system with class-variance-authority, clsx, tailwind-merge

Backend

  • Next.js Route Handlers
  • MongoDB Atlas + Mongoose (profiles + connections)
  • Demo fallback data layer for reliability

Auth

  • Auth0 Universal Login
  • Secure session handling via Auth0 Next.js SDK

Matching Engine

  • Deterministic graph-based scoring system
  • Signals: brokerage, complementarity, serendipity
  • Optional OpenAI (gpt-4o-mini) for explanation enrichment

Challenges we ran into

  • Designing a matching system that feels meaningful, not like generic AI output
  • Making scoring transparent instead of a black-box “AI says so”
  • Ensuring the app works with and without API keys for live demos
  • Balancing strong technical depth with a simple, fast demo experience

Accomplishments that we're proud of

  • Built a full-stack system with Auth0-secured identity and graph-based matching
  • Created an explainable recommendation engine with inspectable signals
  • Designed matches that lead to action (not just suggestions)
  • Delivered a polished UI with clear, research-backed positioning
  • Built a resilient demo system that never breaks during judging

What we learned

  • Weak ties only matter if users understand why the connection is valuable
  • Trust is critical—AI recommendations need transparency and evidence
  • The key metric isn’t match quality—it’s whether users actually reach out

What's next for Zepic

  • Mutual consent before revealing contact details
  • Integration with classes, clubs, and campus events
  • Calendar-based “coffee chat” routing
  • Feedback loops (did this lead to a project, referral, or meeting?)
  • Team formation mode for hackathons and research
  • Privacy controls for AI data usage

Built With

Share this project:

Updates