AegisMesh
AegisMesh is an autonomous, multi-agent AI dispatch swarm which provides a highly concurrent websock
Demo
About
AegisMesh is an autonomous, multi-agent AI dispatch swarm. It provides a highly concurrent WebSocket gateway that civilians can use to broadcast distress signals. The system instantly ingests chaotic text, uses an AI model to extract strict JSON parameters (urgency, disaster type, and exact coordinates), and queries a live spatial index to find the nearest civilian volunteers. It then orchestrates an AI swarm to negotiate the route. If a route is blocked or a volunteer is unavailable, the swarm dynamically loops back, recalculates, and dispatches the next best option—all in milliseconds, broadcasting the live tactical data to a Next.js command center.
How it's made
We architected AegisMesh with an uncompromising focus on low latency and strict data validation: * **The Ingestion Gateway:** Built on **Bun**'s native WebSocket implementation to handle massive connection spikes without blocking the event loop. * **The Geospatial Engine:** Offloaded all distance calculations to **Redis GEO**. Instead of choking the Node event loop with Haversine math, we rely on Redis's C-based, in-memory spatial indexes for sub-millisecond proximity queries. * **The AI Triage:** We utilized **Groq's LPU inference engine** running `llama-3.3-70b-versatile` via the **Vercel AI SDK** to achieve near-instantaneous extraction of coordinates and disaster types from raw text. * **The Zod Fortress:** LLMs hallucinate. We built a strict **Zod** schema firewall to guarantee that no malformed data ever reaches the routing logic. * **The Brain:** We used **LangGraph** to build a cyclical multi-agent graph. This allows the system to recursively search and attempt dispatches until a route is cleared, rather than failing linearly. * **The Frontend:** A reactive **Next.js** App Router dashboard utilizing **Tailwind CSS**, **Shadcn UI**, and **React-Leaflet** for real-time tactical visualization.