cracked.ai

Matching founders with exceptional engineering talent — through AI-powered voice interviews.

The Problem

Europe's best engineers are sitting on the sidelines. Founders who need their first technical hire are burning weeks on LinkedIn, take-home challenges, and leetcode filters that measure the wrong things. The talent exists. The demand exists. The matching layer is broken.

FounderMatch fixes that.

How It Works

For Founders

A 5-minute voice interview replaces the job description. An AI interviewer extracts exactly what you need: required skills, experience level, work arrangement, culture fit, and deal-breakers. No forms. No boilerplate. Just a conversation.

For Candidates

A two-phase voice interview — behavioral first, then a dynamically generated system design question tailored to your actual skills and the real problems matching founders are building. No LeetCode. No generic questions. Real signal.

Matching

After both sides complete their interviews, a deterministic scoring engine computes weighted match scores across skills, experience, culture, and technical performance. Founders control the weights. The dashboard updates in real time.

When a founder finds their match, one click sends a mutual intro email to both parties.

Architecture

Founder Interview (Voice AI) └─ Vapi Web SDK → Vapi Cloud (STT → Claude → TTS) └─ End-of-call webhook → Claude extracts structured profile → Supabase

Candidate Interview — Phase 1: Behavioral (Voice AI) └─ Vapi call → transcript → Claude extracts skills

Between Phases (server-side, ~10s) └─ Claude generates contextual system design question based on candidate skills + matching founder requirements

Candidate Interview — Phase 2: System Design (Voice AI) └─ Generated question baked into system prompt → Vapi call └─ Claude evaluates both transcripts → scores saved to Supabase

Matching Engine (deterministic, no AI) └─ Weighted sub-scores: skills, experience, culture, technical └─ Founder-controlled weights (must sum to 100) └─ Upsert to match_scores → Supabase Realtime → Dashboard

Tech Stack

Layer	Technology
Frontend	Next.js 15 (App Router), TypeScript, Tailwind CSS, shadcn/ui
Backend	Next.js API Routes + Server Actions
Database & Auth	Supabase (Postgres, Auth, RLS, Edge Functions, Realtime)
Voice AI	Vapi.ai (STT, TTS, Claude orchestration via WebRTC)
LLM	Claude (Anthropic) — interviews, extraction, evaluation, question generation
Email	Resend
Deployment	Vercel

Local Setup

Prerequisites

Node.js 18+
A Supabase project
A Vapi account
An Anthropic API key
A Resend account

1. Clone and install

bash git clone cd foundermatch npm install

2. Configure environment variables

bash cp .env.local.example .env.local

env

Supabase

NEXT_PUBLIC_SUPABASE_URL= NEXT_PUBLIC_SUPABASE_ANON_KEY= SUPABASE_SERVICE_ROLE_KEY=

Vapi

VAPI_API_KEY= NEXT_PUBLIC_VAPI_PUBLIC_KEY=

Anthropic

ANTHROPIC_API_KEY=

Resend

RESEND_API_KEY= RESEND_FROM_EMAIL=[email protected]

3. Run database migrations

bash npx supabase db push

Or run supabase/migrations/001_initial_schema.sql directly in the Supabase SQL editor.

4. Start the dev server

bash npm run dev

Open http://localhost:3000.

Key Design Decisions

Why voice interviews instead of forms? Forms produce polished, rehearsed answers. Voice conversations surface how people actually think. Founders articulate needs they wouldn't have written down. Candidates reveal depth (or lack of it) that a resume never would.

Why two-phase candidate interviews? Phase 1 (behavioral) builds a profile. Phase 2 (system design) validates it. The system design question is generated fresh for each candidate based on their stated skills and what matching founders are actually building — so it can't be googled or prepared for.

Why deterministic matching instead of AI? AI is used where it adds value: conducting interviews, extracting structure, evaluating answers. For ranking, transparent math beats a black box. Founders see exactly why a candidate scored how they did, and can adjust weights to reflect what they actually care about.

Why no scores shown to candidates? Candidates who can see their scores will optimize for them. Founders get cleaner signal when candidates are just being themselves.

Project Structure

app/ (auth)/ — login, signup, onboarding (role selection) founder/ interview/ — voice interview UI dashboard/ — ranked candidate cards, connect flow candidate/ interview/ — two-phase voice interview UI api/ interview/ — Vapi end-of-call webhook handlers matching/ — scoring engine trigger + weight updates connect/ — connection creation + intro emails

lib/ supabase/ — client, server, middleware vapi/ — Web SDK wrapper, assistant configs interview/ — prompts, extraction, evaluation, question generation matching/ — deterministic scoring engine email/ — Resend templates (founder intro, candidate intro)

supabase/ migrations/ — initial schema functions/ — match-on-interview edge function