Census

Inspiration

Physical spaces are blind to what people want. In shopping malls, stadiums, and airports, people constantly say what they want — “I want burgers,” “Where’s coffee?” — but no system captures that collective demand as it happens. We wanted a way to listen in real time and make screens respond to the crowd instead of guessing.

What it does

Census is a live demand layer for physical spaces.
It listens to ambient speech, extracts only product and store interest, and ignores personal or private info.
It understands context (e.g., “I want burgers” vs. “I hate burgers”) and tracks the top interests for each location and time window.

Screens update based on what’s most requested in that zone.
If someone says “Fire” or “Emergency,” every screen instantly switches to an alert mode.

How we built it

• Frontend: TypeScript, React, Next.js, TailwindCSS
  
• Speech & AI: Dedalus Labs AI
  
• Code tools: Claude Code
  
• Core flow:
  
  1. Audio sensor captures short chunks
     
  2. Transcription + intent detection
     
  3. Product/store extraction
     
  4. Ranking by zone + time window
     
  5. Real‑time ad screen update + emergency override
     

Challenges we ran into

• Filtering out noise and irrelevant speech
  
• Making the system respond fast enough for live demos
  
• Handling context like “I want X” vs. “I hate X”
  
• Avoiding hallucinations and duplicate detections
  

Accomplishments that we're proud of

• Built a real‑time demand ranking system by zone
  
• Live voice‑triggered emergency mode
  
• Ad screens that update instantly based on top interests
  
• Clean UI that makes the idea easy to explain
  

What we learned

• Real‑time speech systems need strong filtering to avoid false positives
  
• Context matters as much as transcription
  
• Great demos need fast feedback and clear visuals
  

What's next for Census

• More reliable intent detection with larger vocab
  
• Improved accuracy across noisy environments
  
• Pilot tests in real venues (malls, stadiums, airports)
  
• Better ad‑screen personalization per zone

Built With

• dedalus-labs
• figma
• nextjs
• openai
• react
• tailwindcss
• typescript
• vercel
• vite