RDNA AI Guard

The Problem: The Security-Fraud Silo Historically, Information Security (DDoS, IP reputation) and Financial Fraud (velocity, ledger anomalies) operate in separate silos. Attackers exploit this gap through *Smoke Screen Attacks*—launching noisy network DDoS attacks to distract SREs while quietly executing high-value ledger thefts. Traditional systems fail to correlate edge packets with ledger events in real time, and sending sensitive financial logs to public cloud LLMs violates strict compliance frameworks (GDPR, PCI-DSS). The Solution: RDNA AI Guard RDNA AI Guard is a hardened, unified, hybrid-AI FinSec Triage and Threat Routing Engine operating entirely within your secure enterprise perimeter. 1. High-Speed Local AI Triage (Edge Gatekeeper) Ingests multi-vector telemetry packets locally. A preloaded, memory-pinned local LLM (`gemma2:2b` or `gemma4:12b`) performs zero-trust, millisecond-latency triage without sensitive data ever leaving your on-prem servers. 2. Deterministic Parallel Escalation (LangGraph Loop) When high-risk events are flagged, the system triggers an asynchronous cloud-analysis loop using Fireworks AI's `glm-5p2` model. 3. Specialist Agent Cooperation * Our LangGraph architecture fans out to two parallel expert nodes: * CyberSec Agent: Dissects network headers, proxy configurations, malicious IP ranges, and request path vectors. * Anti-Fraud Agent: Analyzes monetary velocity, routing anomalies, structuring patterns, and mule account signatures. 4. Executive Synthesis & Real-Time Streaming * Orchestrator Node: Combines findings into a unified diagnostic report and risk index. * Reactive Sockets Stream: Events persist to PostgreSQL and broadcast live via Server-Sent Events (SSE) to an interactive monitoring dashboard.