Overview
This project is a high-performance CPU-based AI guardrail system designed to detect and block malicious prompts targeting Large Language Models (LLMs).
The architecture was specifically designed for:
Low-latency inference
CPU-only deployment
Enterprise/SOC environments
Air-gapped systems
Scalable inference pipelines
Real-time prompt inspection
The system focuses heavily on:
Prompt injection detection
Jailbreak prevention
Instruction override detection
Data exfiltration prevention
Obfuscation attack detection
Multi-language attack analysis
The guardrail currently supports:
Turkish prompts
English prompts
Mixed Turkish-English prompts
Architecture
The system follows a multi-stage pipeline architecture.
CPU-Based AI Guardrail System
Overview
This project is a high-performance CPU-based AI guardrail system designed to detect and block malicious prompts targeting Large Language Models (LLMs).
The architecture was specifically designed for:
Low-latency inference
CPU-only deployment
Enterprise/SOC environments
Air-gapped systems
Scalable inference pipelines
Real-time prompt inspection
The system focuses heavily on:
Prompt injection detection
Jailbreak prevention
Instruction override detection
Data exfiltration prevention
Obfuscation attack detection
Multi-language attack analysis
The guardrail currently supports:
Turkish prompts
English prompts
Mixed Turkish-English prompts
Architecture
The system follows a multi-stage pipeline architecture.
Detection Pipeline
1. Language Detection
The first stage identifies the language of the incoming pr…
