Full Stack AI MVP | Prompt-to-Design, API Benchmarking | 5 Days

Built for Josh Zipkowitz as a rapid-turnaround MVP engagement. The goal: a production-grade AI tool that generates premium event invitation images from natural language prompts or structured form inputs, with real-time API benchmarking across multiple image generation providers.

Event planners and invitation designers spend hours manually creating custom invitation graphics. Existing tools are either too generic (Canva templates) or too complex (Photoshop from scratch). The client needed an AI-powered solution that could generate unique, high-quality invitation images in seconds, with enough control to match specific event themes, color palettes, and styles.

Prompt-to-Design Pipeline: Users describe their event in natural language ("elegant garden wedding with gold accents and floral borders") and the system generates multiple invitation design options in under 30 seconds.

Smart Form System: For users who prefer guided input over freeform prompts, I built a structured form flow that captures event type, color scheme, style preferences, text content, and layout options, then converts these into optimized prompts behind the scenes.

Multi-API Benchmarking Engine: The system evaluates image generation results across multiple AI providers (including OpenAI DALL-E and Stability AI), comparing output quality, generation speed, cost per image, and style consistency. This gives the client data-driven insight into which API performs best for different invitation styles.

Full-Stack Web Application: Clean, responsive React frontend with Next.js for server-side rendering and API routes. Supabase handles auth and data persistence. Deployed on Vercel for instant global access.

Frontend: React, Next.js, Tailwind CSS

Backend: Next.js API routes, Python microservices for image processing

AI / ML: OpenAI API ( DALL-E ), Stability AI API, custom prompt engineering

Database: Supabase (PostgreSQL)

Deployment: Vercel

API Integration: RESTful architecture with rate limiting, error handling, and fallback logic across multiple AI providers

Scoped, built, and shipped in 5 days ( Jan 1-5, 2026 ). The MVP was delivered with full documentation, API benchmarking reports, and a clean codebase ready for iteration.