AgentShield

Python, TypeScript, PyTorch, FastAPI, LangChain, Docker. Focus on ML security, transformer classifiers, and API design.

Adversarial ML, prompt injection patterns, transformer fine-tuning for security classification. Also exploring eBPF for runtime agent monitoring.

AgentShield — open-source prompt injection detection for LLM agents. Fine-tuned DeBERTa classifier, sub-3ms latency, F1 0.921 on public benchmarks.

Conversations about LLM security, AI agent hardening, and prompt injection defense. Open to integrations and collaborations around agent safety.