Architecture Specification v3.0

SENTINEL SV3

Cybersecurity Threat Detection System — 4-layer architecture from DNS perimeter to AI observability.

The only system covering physical, network, and AI-application security simultaneously.

Status: IN BUILD · NAICS 541519 · NSF SaTC Narrative Ready

Four-Layer Architecture

Each layer addresses a distinct threat surface. Together they form a unified physical-to-AI security monitoring system.

L0

Pi-Hole DNS Perimeter

Deploy this week

Network DNS Filtering Layer

Network-wide DNS blocking. First line of defense — blocks known malicious domains before they reach any device on the network.

L1

Polariton Detection

Partnership pending

Quantum Photonic Sensing Layer

Quantum photonic sensing for physical-layer threat detection. Partnership with CCNY / Prof. Vinod Menon (pending confirmation).

L2

Hailo-10H Neural Classification

Hardware ordered

Edge AI Threat Classification

Edge AI processing at 40 TOPS INT4 on 2.5W power. Offline-capable neural network classification of network traffic patterns.

L3

SENTINEL-AI (LLM Observability)

In development

AI Session Security Monitor

Monitors RECA Agent sessions for MITRE ATLAS threats. Detects prompt injection, RAG poisoning, and LLM jacking in real-time.

Hardware Stack

Raspberry Pi 5

$80

Hailo-10H HAT+ 2

$130

Accessories

~$20

Total: ~$230

Deployable this week

Threat Coverage Matrix

Threat VectorDetection LayerMITRE Reference
DNS C2 CommunicationLayer 0T1071.004
Credential Access / PhishingLayer 0T1566
Physical RF InjectionLayer 1Physical
Electromagnetic InterferenceLayer 1Physical
Network Traffic AnomaliesLayer 2T1040
Prompt InjectionLayer 3ATLAS T0035
LLM JackingLayer 3ATLAS T0051
RAG PoisoningLayer 3ATLAS T0054
Supply Chain (Model Weights)Layer 3T1195

RECA Anti-Poison Defense

SENTINEL detects threats from outside. RECA Anti-Poison defends from inside. Together they form a complete AI security perimeter.

SENTINEL (External)

Monitors network traffic, DNS queries, and AI session patterns for known threat signatures.

Anti-Poison (Internal)

5-layer defense within RECA Agent sessions: input validation, context verification, output filtering, session integrity, audit logging.

Grant & Procurement Narrative

NSF SaTC Fit

First unified physical-to-AI-application security monitoring system. Novel contribution: polariton-based physical threat detection integrated with LLM observability.

NAICS Codes

541519 (Computer Systems Design) · 334519 (Electronic Instruments)

Fiscal Sponsor: The School of Arts and Performing Arts Inc. · EIN 47-3710795