SiteSignal CALL-E

When a human has no signal, CALL-E reaches out to them. Voice-first incident coordination for environments where internet and SMS don't reach.

Comment

Inspiration

Maintenance workers often operate in dead zones — elevator shafts, basements, utility tunnels. When something goes wrong, they can't open a ticket, send a message, or search for help. We asked: what if the system called them instead?

What it does

SiteSignal CALL-E is an operator coordination platform that closes the loop for offline workers. When an incident is reported, the operator dispatches CALL-E to call the worker directly. CALL-E collects status, identifies what resource is needed, finds nearby suppliers, and calls the worker back with options. The operator then dispatches a runner to deliver the item — all without the worker needing internet access.

How we built it

  • Backend: Node.js + TypeScript + Fastify, deployed on Google Cloud Run
  • Voice AI: CALL-E API for outbound calls with structured result extraction
  • Storage: Google Firestore for real-time incident state
  • Frontend: React + Three.js for a 3D campus map with live incident queue
  • Infrastructure: Google Cloud Run, Secret Manager, Cloud Build

Challenges we ran into

  • CALL-E call screening: carriers intercept the call before the human answers, causing the agent to start mid-script. We solved this by instructing the agent to pause after screening and wait for a human greeting.
  • Idempotency key collisions on demo resets: fixed by generating a session ID per incident that factors into the idempotency key.
  • Structured data extraction: getting CALL-E to reliably capture item, ZIP code, and supplier preference from natural conversation required careful prompt design.

Accomplishments that we're proud of

  • A fully closed loop: report → CALL-E calls worker → operator selects supplier → CALL-E calls back → runner dispatched
  • Zero internet required on the worker's side — voice only
  • Live demo with a real phone call during the submission

What we learned

Voice AI agents need clear context about who they're talking to. A maintenance technician in an elevator shaft is doing their job — not an emergency. Prompt framing determines whether the agent panics or coordinates.

What's next for SiteSignal CALL-E

  • Operator supplier selection with real-time inventory lookup
  • Multi-site support with zone-aware routing
  • Integration with facility management systems (CMMS)
  • Support for non-English workers

Built With

Share this project:

Updates

Submission history