bryand410/llm-sentinel

# 🛡️ LLM Sentinel An ultra-fast, local AI gateway designed to detect and block Prompt Injections and Jailbreaks before they hit your massive LLMs (ChatGPT, Claude, etc.). ## 🧠 Architecture - **Model:** Fine-tuned `prajjwal1/bert-tiny` for sequence classification. - **API:** Asynchronous FastAPI backend. - **Latency:** < 50ms inference time on CPU. ## 🚀 Quick Start 1. `pip install -r requirements.txt` 2. Train the sentinel: `python src/train.py` 3. Launch the API: `uvicorn api.main:app --reload`