koveri

AI-powered social discovery for meeting people with shared taste, projects, and context.

koveri builds a compatibility profile from a user's online presence — GitHub, Spotify, Letterboxd, Twitter/X, Substack, Steam, and optional photos — then turns that data into a swipeable discovery experience, messaging flow, and AI coach for profile improvements.

Originally prototyped as gemini-connect; renamed and polished as koveri.

Why it exists

Most social apps make people describe themselves from scratch. koveri explores a different flow: use public digital traces to surface concrete conversation starters like repos, music taste, film history, interests, and creative work.

The app can be aimed at different discovery goals, including:

• finding friends
• meeting potential dates
• finding hackathon or project collaborators

Features

• Guided onboarding for basic info, platform usernames, profile goals, and photos.
• Online-presence extraction through a Supabase Edge Function proxy for Yellowcake.
• AI profile generation that turns extracted data into bios, prompt answers, highlights, fun facts, and data-backed insights.
• Gemini-style profile coach for conversational profile edits and optimization.
• Swipeable discovery feed with compatibility cards, prompts, match badges, and like/skip interactions.
• Local messaging prototype for starting conversations with discovered profiles.
• Local-first demo state using localStorage/IndexedDB for profile, matches, messages, and photos.

Tech stack

Frontend

• React 18
• TypeScript
• Vite
• Tailwind CSS + shadcn/ui
• Framer Motion
• Recharts
• Vitest

Backend / AI services

• Supabase Edge Functions
• Yellowcake API for web extraction
• Lovable AI Gateway / Gemini model for profile generation and coaching

Architecture

User onboarding
  ├─ platform usernames + photos + discovery goal
  ↓
Yellowcake extraction proxy
  ├─ GitHub repos
  ├─ Spotify playlists/tracks
  ├─ Letterboxd films
  ├─ Twitter/X posts
  ├─ Substack posts
  └─ Steam games
  ↓
AI profile generation
  ├─ bio
  ├─ prompt answers
  ├─ fun facts
  ├─ data insights
  └─ best features
  ↓
Discovery + profile coach + messages

The frontend talks to Supabase Edge Functions instead of calling private APIs directly from the browser:

• supabase/functions/yellowcake-proxy keeps YELLOWCAKE_API_KEY server-side.
• supabase/functions/generate-profile generates profile content.
• supabase/functions/coach-chat powers the profile coaching chat.

Getting started

Prerequisites

• Node.js 18+
• npm
• A Supabase project if you want to run the Edge Function integrations

Install

npm install

Configure environment

Copy the example file:

cp .env.example .env

Set the frontend Supabase values:

VITE_SUPABASE_URL=your_supabase_project_url
VITE_SUPABASE_PUBLISHABLE_KEY=your_supabase_publishable_key

Server-side secrets should be configured in Supabase, not committed to this repo:

supabase secrets set YELLOWCAKE_API_KEY=...
supabase secrets set LOVABLE_API_KEY=...

Run locally

npm run dev

Build

npm run build

Test

npm test

Lint

npm run lint

Security notes

• .env is intentionally ignored and should never be committed.
• API keys for Yellowcake and AI providers belong in Supabase Edge Function secrets.
• This is a prototype; any production version should add authentication, database-backed authorization, rate limits, and privacy controls around extracted profile data.

Project status

Prototype / portfolio project. The core product flow is implemented, but some data is still mocked or demo-oriented. The next polish pass would add real auth, production persistence, more tests, and deployment docs.

My role

Solo implementation of the React/Vite frontend, onboarding flow, local data persistence, discovery UI, AI prompt system, profile coach, Supabase Edge Function integrations, and Yellowcake extraction flow.

Future improvements

• Add authentication and real user-backed Supabase persistence.
• Replace remaining mock discovery data with generated or database-backed profiles.
• Add screenshots, a hosted demo link, and a short product walkthrough video.
• Add unit/integration tests for profile generation, local storage hooks, and messaging flows.
• Add privacy controls for imported data and user deletion/export flows.