The Challenge
Problem
Restoring damaged photos usually requires expensive software (Photoshop) and deep technical expertise. Existing AI tools are often behind paywalls or require complex local setups.
Solution
A browser-based platform where users can upload images and apply restoration pipelines with a single click. The backend handles the heavy lifting using pre-trained GFPGAN and DeOldify models served via a Flask API.
Engineering Hurdles
Handling Large Image Payloads
High-res images caused timeout issues on the free-tier backend. Solved by implementing chunked uploads and asynchronous processing queues with Celery.
Model Inference Latency
Initial model loading took 10+ seconds. Optimization involved model quantization and keeping warm instances during peak usage.
Key Features
Tech Stack
frontend
React, Tailwind CSS, Framer Motion
backend
Python, Flask, OpenCV
ml
PyTorch, GFPGAN, DeOldify
deployment
Vercel (Frontend), Render (Backend)
My Role
Full Stack Engineer