Fast Gradient Sign Method adversarial attacks on ML models with both local and cloud deployment paths.
Project Overview
This project is an adversarial machine learning demo built around the Fast Gradient Sign Method. It supports local testing and cloud deployment through AWS Lambda, Amplify, ECR, and a React frontend.
What I Built
- Local FastAPI backend for testing
- Cloud deployment flow with AWS services
- React-based interface for experiments
- Practical adversarial ML workflow
Problem
Security risks in machine learning are often explained academically, but developers rarely get a usable demo that shows how adversarial attacks behave in practice and how to expose them safely.
Solution
This project packages FGSM attack logic into a deployable product with an API layer, frontend, and cloud infrastructure so the concept becomes much more tangible than a notebook-only demo.
Why It Matters
It makes adversarial ML easier to teach, test, and demonstrate while also showing deployment discipline beyond the research stage.
Stack
- Python
- FastAPI
- AWS
- React
What I Learned
This build showed me how quickly ML security work becomes a systems and product problem, not just a model problem.
- research ideas need better tooling to become understandable
- deployment paths change how security experiments feel in practice
- frontend visibility makes technical concepts easier to explore