HACKATHON PRESENTATION • SLIDE 01 / 09
LIVE
⚔️ HACKATHON 2026 — AI TRACK
WAR
ROOM
An immersive, multi-agent AI crisis simulation platform where you command a room full of specialist AI advisors — each with their own voice, expertise, and agenda — to navigate high-stakes, real-time crisis scenarios.
🔴 THREAT LEVEL: CRITICAL
01 / 09
THE PROBLEM
LIVE
⚡ PROBLEM STATEMENT
CRISIS DECISIONS ARE MADE BLIND
🏢
Enterprise
No Safe Practice Ground
Organizations drill crisis scenarios with paper exercises. When a real breach hits, leadership has never actually felt the pressure of a live crisis room.
🧠
Consulting
Expensive Simulations
Traditional tabletop exercises cost thousands, require human role-players, and can't realistically simulate rapidly escalating, multi-stakeholder crises.
🎙️
The Missing Frontier
Voice + Multi-Agent = Unsolved
Voice AI exists. Multi-agent AI exists. Nobody has fused them into one coherent, real-time experience where AI advisors speak, debate, and respond to you — live, without a script.
02 / 09
THE SOLUTION
LIVE
✅ WAR ROOM — THE SOLUTION
YOUR PERSONAL AI CRISIS COMMAND CENTER
Enter a crisis scenario. A team of AI specialists assembles around you — each one speaks in their own voice, debates in real time, and pushes back on your decisions. You are the Chairman. Every call is yours.
🎭
Dynamically Generated Teams
The Scenario Analyst agent reads your crisis brief and assembles the exact specialists you need — Legal, PR, Engineering, Finance — each with tailored expertise.
🎙️
Real-Time Voice Debate
Every agent speaks aloud using ElevenLabs ultra-realistic TTS. They interrupt, they agree, they conflict. The room feels alive because it is.
🌍
Escalating World Events
The World Agent injects breaking news and new developments on a timer. Nothing stays static. Your crisis evolves whether you're ready or not.
03 / 09
HOW IT WORKS
LIVE
⚙️ AGENT FLOW — FROM SCENARIO TO SESSION
THE AGENT ASSEMBLY CHAIN
SCENARIO ANALYST — The Architect
Reads your crisis brief → decides which specialists are needed → spawns and configures each sub-agent with a unique name, role, voice, and expertise
SPAWNS TEAM
SPECIALIST ADVISORS — The War Council
e.g. Legal Expert, Chief Communications Officer, Engineering Lead — each runs an independent reasoning loop, speaks in voice, and posts decisions to the Crisis Board
LIVE DEBATE
WORLD AGENT — The Game Master
Runs on a timer → injects escalation events, breaking news, and new complications into the session regardless of your current decisions
ESCALATES
OBSERVER AGENT — The Scorekeeper
Silently monitors all transcripts → evaluates decisions → updates live trust scores, public confidence, and crisis posture metrics in real time
SCORES
04 / 09
TECH STACK
LIVE
🔧 TECHNOLOGY STACK
BUILT WITH REAL INFRASTRUCTURE
INTELLIGENCE LAYER
🤖
Z.AI GLM Models
Primary LLM for agent reasoning via OpenAI-compatible SDK
🎙️
ElevenLabs
Ultra-realistic TTS/STT — each agent has a unique voice
📡
LiveKit (WebRTC)
Real-time bidirectional audio streaming pipeline
PLATFORM LAYER
⚡
FastAPI (Python)
Async backend gateway — REST + WebSocket orchestration
⚛️
Next.js (React)
Cinematic tactical UI with real-time state sync
🔥
Firestore
Event-driven state store with snapshot listeners
05 / 09
SYSTEM ARCHITECTURE
LIVE
🏗️ ARCHITECTURE OVERVIEW
THREE PIPELINES. ONE LIVE ROOM.
Pipeline 01 — State
📊
REST + WebSocket Event Bus
Frontend → GET /api/sessions
Firestore → Snapshot Listener
Gateway → Push via ws://

Every agent action instantly syncs to the UI. No polling. No lag.
Pipeline 02 — Voice
🎙️
Bidirectional Audio Stream
Chairman Mic → LiveKit
ElevenLabs STT → Transcript
Z.AI GLM → Agent Response
ElevenLabs TTS → Audio Out
Pipeline 03 — Agent Isolation
🔒
Isolated Memory + Shared Board
Each agent has private memory. They share only the Crisis Board. No agent can read another's internal reasoning — just like real advisors in separate rooms.
06 / 09
WHAT WE BUILT
LIVE
📐 SCOPE & COMPLEXITY
WHY THIS IS NON-TRIVIAL
5+
Concurrent AI Agents per Session
3
Real-Time Pipelines (State, Audio, Events)
∞
Unique Scenarios — Dynamically Generated
0ms
UI Polling — Pure Event-Driven Push
Hard Engineering Problem
Active Speaker Gating
We solved multi-agent voice leakage — where multiple AIs would talk over each other — with strict, backend-enforced speaking locks and token-based turn management.
Hard AI Problem
Prompt Architecture at Scale
Each agent maintains isolated long-term memory while sharing a global crisis state. Custom base agent class with tool-based shared board read/write access.
07 / 09
APPLICATIONS & VISION
LIVE
🚀 WHO NEEDS THIS
REAL WORLD APPLICATIONS
🏢
Enterprise Training
C-Suite Crisis Readiness
Boards and executive teams use WAR ROOM to rehearse cyberattacks, regulatory crises, PR disasters, and supply chain failures — before they happen for real.
🎓
Education
Business School Simulations
MBA programs replace expensive human role-playing exercises with unlimited, AI-driven scenario simulations accessible from a browser.
🛡️
Government & Defense
Policy Stress-Testing
Government agencies simulate policy decisions under crisis conditions — pandemic response, infrastructure failures, geopolitical escalations.
08 / 09
CLOSING
LIVE
⚔️ THE CALL TO ACTION
THE ROOM IS WAITING
WHAT WE DEMONSTRATED TODAY
01
Multi-Agent AI Orchestration
Multiple specialist AI agents, each with memory and voice, debating a real crisis in real time
02
Dynamic Scenario Generation
The Scenario Analyst assembles a bespoke expert team for every unique crisis — no two sessions are the same
03
Full-Stack Real-Time Infrastructure
Event-driven architecture across voice, state, and AI decision pipelines with zero polling
✓
Built & Working — Right Now
This is not a mockup. WAR ROOM is a deployed, functional system.
09 / 09