View on GitHub
Frogstagram is Instagram with one key difference: it only allows photos of frogs. Using a YOLOv11 computer vision model, the platform automatically filters uploads to maintain a strictly frog-focused feed. This creates a unique use case for deploying ML in production while serving a specific community.
Core Technical Features:
- Frontend built with SvelteKit and Tailwind CSS for responsive, modern UI
- Backend Service:
- FastAPI running on AWS Lambda for efficient API handling
- Computer vision using YOLOv11 model in Docker containers via Amazon ECR
- Resulted in automated image filtering ensuring frog-only content
- Other AWS Infrastructure:
- S3 for scalable image storage and delivery
- Cognito handling user authentication
When users upload a photo, it’s sent to a Lambda function running the image classification model. The model determines if the image contains a frog - if it does, the image is stored in S3 and appears in feeds. If not, it’s rejected. This process ensures the feed remains only frog photos.
Frogstagram feed page |
Upload interface with ML-powered verification |
Live Demo
Note: The legacy version of this project, which used vanilla JavaScript for the frontend and FastAPI with SQLAlchemy for the backend, and TensorFlow for inference, can be found at Frogstagram Legacy. The current production version is at Frogstagram Live.