MarketMate — AI Marketing Tool

The Problem

NY TechWeek Hackathon: build a product that solves a real marketing problem using AI, in 48 hours, with a working demo and live users at presentation.

The team identified a common pain point: small business owners spend hours writing marketing copy — emails, social posts, product descriptions — and iterate manually without knowing what works. We built a tool that generates multiple content variants and tracks which ones get used.

What Was Built

• Content generation engine: OpenAI API integration with per-content-type prompt templates (email subject lines, social captions, product descriptions). Each type returns 3 variants for comparison.
• Usage tracking: Supabase logs which variants users select, feeding a simple analytics dashboard.
• Authentication: Supabase Auth with email/password and session persistence.
• Vercel deployment: pushed at hour 20 (not hour 47) to allow real external user testing before the demo.

Architecture Decisions

Structured prompting vs free-form Free-form generation produced inconsistent output. Switched to per-content-type templates with explicit format instructions (e.g., "Generate 3 email subject lines under 60 characters each, using action verbs"). Consistency improved dramatically.

Deploy early Deploying at hour 20 instead of hour 47 gave 28 hours to test with real users, catch environment-specific bugs, and collect actual usage data to show judges.

Process

• Hours 0–2: Defined MVP scope. One content type end-to-end, with auth, generation, selection, and tracking. Everything else was stretch.
• Hours 2–14: Divided work — auth + DB schema / OpenAI integration + prompt design / frontend interface. Synced every 3 hours.
• Hours 14–20: Integration + QA. Fixed auth token passing between frontend and API routes.
• Hour 20: First Vercel deploy. Fixed one environment variable misconfiguration.
• Hours 20–44: Added two additional content types, shared URL with 8 external testers, made 3 UX adjustments based on feedback.
• Hours 44–48: Polish and demo prep.

Results

• Working product shipped in 48 hours — all core features functional at demo time
• Live demo with real usage data: 8 external users, 43 content generations tracked
• Secure production deployment on Vercel with proper environment variable management
• Consistent AI output quality achieved through per-content-type prompt templates
• Team of 3 coordinated across frontend, backend, and AI integration without overlap