ML Engineering · MLOps · Code Security

Out of Harm's Way

175K+ SAMPLES31 CWEs2-STAGE ML PIPELINE

The Hook

AI that finds vulnerabilities — and attempts to fix them.

The Challenge

Code vulnerabilities are hard to detect manually across 31 CWE classes, and even harder to auto-remediate at scale.

What We Built

An end-to-end MLOps research system: a GraphCodeBERT multi-label classifier + CodeT5+ fix generator, trained on 175K+ samples with a Gradio demo.

Role

ML Engineering

MLOps

Model Fine-tuning

API Deployment

Technologies

PyTorchGraphCodeBERTCodeT5+FastAPIHugging Face