Objectives

1. Build an end-to-end AI image generation platform from frontend to backend architecture.
2. Optimize serverless GPU cold starts for a seamless user experience.
3. Create brand identity for “PhotoFiltr”.

Front-end & Design

Designed a “neobrutalistic” playful UI. The frontend includes a custom user dashboard, Google OAuth for sign-ups, and a Stripe-powered token system for image processing. To manage storage costs and user privacy, I implemented Cloudflare R2 buckets with automated lifecycle policies to delete user data after a set period.

Technical Architecture & AI

‍The core of the project involved porting a local workflow into a scalable cloud environment:

1. Backend: Engineered a custom Python backend with SQL to manage user profiles, credits, and job history.
2. Infrastructure: Hosted the AI model on Modal, a serverless GPU platform, which I configured to handle image manipulation requests.
3. Performance:  Reduced container cold starts to 2 seconds, ensuring users aren’t left waiting for the GPU to spin up.

Key Contributions

• Logo Design: Created the visual brand identity for Photofiltr.
• Full-Stack Development: Managed the entire stack from Webflow frontend, to the SQL database and to the serverless Python environment.
• Infrastructure Management: Handled the deployment and optimization of the AI model on Modal.