I recently built and deployed a real-time fraud detection API using GraphSAGE GNNs — live at fraudguardapi.xyz. Here's what I learned.
Why GraphSAGE?
Regular fraud detection models look at each transaction in isolation — just amount, time, merchant. Mine looks at the entire network of connected accounts. Fraud rings only show up when you look at the graph structure, not individual transactions.
The stack:
GraphSAGE model (3-layer, 97.6% AUC)
FastAPI backend
AWS EC2 Mumbai region
Nginx reverse proxy + Let's Encrypt SSL
React landing page
Hardest bug I hit:
The live site kept serving a stale JS bundle despite redeploying. Spent hours on it. Traced it to nested scp upload paths overwriting the wrong directory. Fixed by wiping the destination folder completely and recopying from the correct build path.
What's next:
I will be adding Docker + CI/CD pipeline so this never happens again.
If you're building something in fintech or need an ML model deployed as a production API — I'm open to freelance projects. DM me or drop a comment.
Live demo: fraudguardapi.xyz
Top comments (0)