PULSE v1.0.0 — The AI Nervous System for Smart Stadiums

PULSE is an agentic AI operating system for live-event safety, built for PromptWars Challenge 4. It is not a chatbot: four named agents run a continuous observe → reason → decide → act → verify loop over a live simulated 62,000-seat stadium, and every decision is explained and written to an auditable ledger.

This first release is end-to-end functional in Demo Mode with zero configuration — the console auto-starts a live simulation and a deterministic heuristic engine drives all four agents. Add a GEMINI_API_KEY to upgrade the reasoning to the real model, and Firebase keys to enable real auth and persistence.

The four agents

• Sentinel (Vision) — per-zone density perception and a 15-minute risk forecast with pre-emptive alerts.
• Strategist (Brain) — triages every incident to a 1–5 severity, ranks priorities, and renders a step-by-step reasoning chain.
• Marshal (Dispatch) — assigns the nearest qualified responder on the zone graph, with a live SLA timer and lifecycle tracking.
• Guardian (Reunite) — extracts a privacy-safe descriptor from a lost-child report and scores it against steward sightings. No biometrics, no images, human-confirmed.

Highlights

• Six Gemini route handlers, each with strict structured output, Zod re-validation, and a deterministic heuristic fallback in the identical shape.
• Animated SVG stadium digital twin, decision ledger, analytics with CSV/JSON export, command palette, and a dark-first design system.
• Content-Security-Policy and a full set of response security headers.

Verification

npm run build passes, tsc --noEmit is clean, and ESLint reports 0 errors (a few documented React-Compiler-preview advisory warnings).

See the README, architecture, and demo guide.