Echo - The Infinite Context City Brain

The memory layer for sentient cities. 10x log compression with Token Company bear-1 so the City Brain remembers 7 days instead of 5 minutes. 67% faster emergency response.

Comment

Inspiration

The hackathon theme — "Turing City: sentient digital metropolis" — sparked a critical question: what makes a city truly sentient? It's not just having AI agents for traffic, power, and emergencies. It's having them work together with shared memory.

The problem? Context windows. When the City Brain tries to listen to every agent at once, it runs out of tokens. The city develops amnesia — it literally forgets what happened 10 minutes ago. Critical patterns are lost. Emergency response slows down.

Without Echo, the City Brain is stateless. With Echo, it becomes stateful.

What it does

Echo is the subconscious memory layer for sentient cities. It:

  1. Ingests high-volume logs from multiple city AI agents (Traffic, Power, Emergency, Security)
  2. Compresses them 10x using Token Company's bear-1 model while preserving 95%+ semantic meaning
  3. Enables the City Brain (powered by Google Gemini) to access 7 DAYS of historical context instead of just 5 minutes
  4. Speaks critical decisions aloud using ElevenLabs voice synthesis
  5. Monitors everything with Arize-ready observability architecture

The Demo: Fire in Sector 7

Without Echo With Echo
City Brain sees only 5 minutes of logs City Brain has 7 DAYS of context
Doesn't know Fire Truck 3 was dispatched Knows all truck positions & traffic
Sends wrong truck → 15-min delay Routes nearest truck optimally
Response time: 18 minutes Response time: 6 minutes (67% faster)

How I built it

  • Frontend: Next.js 14 with React 19, TypeScript, and custom cyberpunk-themed UI
  • Compression: Token Company bear-1 API for 10x context reduction with intelligent fallback
  • City Brain: Google Gemini API for real-time intelligent decision-making
  • Voice Alerts: ElevenLabs TTS so the City Brain speaks emergency decisions
  • Observability: Arize-ready architecture for monitoring AI decisions
  • Visualization: Recharts for real-time data display, Framer Motion for animations

Challenges I ran into

  • Generating realistic multi-agent logs that feel authentic and varied
  • Balancing compression ratio with semantic preservation — needed 10x compression while keeping critical entities
  • Building an immersive UI that feels like a real city command center
  • Solo-building a complex multi-system architecture under time pressure

Accomplishments that I'm proud of

  • Achieved 10x compression while maintaining 95%+ semantic retention
  • Built a fully functioning 4-agent simulation system with realistic city scenarios
  • Created an immersive cyberpunk command center experience that's visually stunning
  • As a solo hacker, shipped a complete end-to-end system with voice, compression, and AI decision-making

What I learned

  • The power of LLM input compression for extending context windows cost-effectively
  • How to design multi-agent systems with shared memory architecture
  • The importance of observability in AI-powered critical infrastructure

What's next for Echo

  • Integration with real city IoT sensors and traffic APIs
  • Expansion to specialized agents: Healthcare, Education, Finance
  • Enterprise deployment for smart city initiatives worldwide
  • Open-source release for the developer community

Built With

Share this project:

Updates