AegisMesh

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.

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.