Crisis Core: High-Stakes Conversation Simulator

🧠 Core Functionality

Crisis Core is an immersive 911 dispatcher training simulator that places users in realistic emergency call scenarios. The application orchestrates five distinct AI caller agents, each powered by ElevenLabs Conversational AI, presenting unique emergency situations that require critical thinking and rapid information extraction.

The core interaction loop:

• Users receive incoming voice calls from AI-powered “callers” in distress.
• Through real-time two-way voice communication, users must extract critical details: caller name, location, situation severity, and emergency type.
• Users complete a dispatch report, selecting appropriate response units (Ambulance, Police, Fire Dept, Animal Control).
• An AI evaluator scores performance against the actual conversation transcript.

---

✅ Working Prototype Stability

The prototype demonstrates production-ready stability:

• Seamless voice session management with connection/disconnection handling.
• Persistent session state via sessionStorage to track progress across five calls.
• Graceful error handling for microphone permissions and API failures.
• Real-time call duration tracking and mute functionality.
• Automatic call progression with “Connecting…” transition states.

---

⚙️ Technical Complexity

🔀 Multimodal Orchestration

• Voice Input/Output: Real-time bidirectional audio streaming via ElevenLabs WebRTC.
• Text Processing: Transcript extraction from conversation history.
• AI Evaluation: GPT-4o-mini analyzes transcript against user-submitted dispatch reports.
• State Machine: Complex game state management (idle → inCall → reportPending → evaluating → result → complete).

🔧 Tool Integration

• Browser microphone access via MediaDevices API.
• Serverless function orchestration for secure API calls.
• Cross-service data flow:
  ElevenLabs conversation → transcript retrieval → OpenAI evaluation → scored feedback

---

🌟 Innovation & Creativity

Crisis Core transforms passive learning into active, pressure-tested skill development. Unlike traditional training materials, users experience the emotional intensity of real emergency calls — where callers may be panicked, confused, or uncooperative.

The AI agents create dynamic, unpredictable scenarios that can’t be memorized, forcing genuine skill application.

The evaluation system provides objective, consistent feedback that is impossible to scale with human trainers. Transcripts enable self-reflection and help users improve their communication patterns.

---

🌍 Real-World Impact

Emergency dispatch training is critical — but expensive. Traditional methods require:

• Experienced dispatchers taken off active duty to train.
• Scripted role-play scenarios that quickly become predictable.
• Subjective evaluation that varies by trainer.

Crisis Core enables:

• Scalable training for dispatch centers of any size.
• 24/7 availability for self-paced learning.
• Consistent evaluation standards across all trainees.
• Safe failure environment where mistakes are learning moments.
• Reduced training costs while increasing practice volume.

Future expansion possibilities include hostage negotiation training, courtroom debate practice, and medical triage simulations.

---

🎯 Theme Alignment: Cohesive Agent Orchestration

Component	Role in Agent Ecosystem
Browser	Hosts the interactive dispatch console, manages audio I/O, maintains session state
Voice (ElevenLabs)	5 distinct caller agents with unique personalities and scenarios, real-time speech synthesis and recognition
Cloud (Netlify Functions)	Serverless orchestration layer connecting ElevenLabs transcript API to OpenAI evaluation
Tools (OpenAI GPT-5-nano)	Intelligent evaluation agent analyzing dispatch accuracy against ground truth

The browser acts as the command center, the voice agents simulate realistic human interaction, the cloud provides secure orchestration, and the LLM delivers intelligent assessment — all working together as a unified training system.

---

🛠 Technology Stack

Category	Technology
Voice AI	ElevenLabs Conversational AI SDK (@elevenlabs/react v0.12.1)
Evaluation LLM	OpenAI GPT-5-nano
Frontend Framework	React 19
Build Tool	Vite 7
Serverless Backend	Netlify Functions
Hosting	Netlify
APIs	ElevenLabs Conversations API, OpenAI Chat Completions API
State Management	React hooks + sessionStorage persistence
Styling	Custom CSS with CSS variables, Google Fonts (Inter, JetBrains Mono)
Audio	Browser MediaDevices API, WebRTC (via ElevenLabs SDK)