Inspiration

Business school gives you access to extraordinary people — but no one has time to figure out who to meet, why, or when. We watched our MBA cohort of 130+ professionals struggle with the same problem: too many brilliant people, too little surface area. Introductions happened by accident, not design.

We asked: What if your cohort had a concierge — one that knew everyone, remembered everything, and never slept?

What it does

Alma is an AI-powered private network intelligence platform for Columbia EMBA cohorts. It turns a directory of classmates into a living, breathing network through:

  • AI Concierge — a conversational assistant that knows every member's background, interests, and availability. Ask it anything: "Who should I grab dinner with Thursday?" or "Who here has fintech experience?"
  • Alma Match — intent-based AI matchmaking. Describe what you're looking for and get ranked recommendations with shared-context reasoning.
  • Blackboard — a unified feed of events, chronicles (shared memories), gratitude notes, and needs — keeping the cohort connected between residencies.
  • Event Prep / Placemats — AI-generated intel briefs on attendees before any gathering, powered by LinkUp web research.
  • Network Vitality Score — a real-time pulse on how actively you're engaging your network.

How we built it

  • Frontend: React Native + Expo (cross-platform: iOS, Android, Web) with a custom design system — Inter typography, pure-black surfaces, editorial layout inspired by NYT meets X.
  • Backend: Supabase (PostgreSQL + Auth + Realtime + Edge Functions) for the entire data layer.
  • AI: Claude API powering the concierge chat, matchmaking engine, and placemat generation. LinkUp API for real-time web intelligence on attendees.
  • Design: Zero third-party UI libraries — every component hand-built for brand coherence. Dark theme with champagne-gold accents, glassmorphism, and gradient CTAs.

Challenges we faced

  • Cold start problem: Making the AI concierge useful from day one required pre-seeding member profiles with LinkedIn data, identity tags, and enrichment.
  • Cross-platform consistency: Getting LinearGradient, backdrop blur, and custom fonts to render identically on web and native required platform-specific conditionals throughout.
  • Real-time at scale: Supabase Realtime channels for live notifications, NVS updates, and chat streaming needed careful subscription management to avoid memory leaks.
  • Design at speed: We rebuilt the entire UI twice in 48 hours — from generic dark theme to a premium editorial aesthetic — touching 50+ files with consistent design tokens.

What we learned

The hardest networking problem isn't access — it's activation. People don't need more contacts; they need the right nudge at the right time. AI isn't replacing human connection here — it's removing the friction that prevents it.

Built With

Share this project:

Updates