E-mploye for CALL-E

What I built

E-mploye is one virtual employee for routine business calls. A manager selects a task, previews the exact instruction, requests approval, authorizes the call, reviews the status and evidence, and decides whether the result can be applied.

The MVP includes three business contexts:

• Appointment desk: confirm or reschedule a customer appointment.

• Lead follow-up: agree on a useful follow-up time with a prospect.

• Shift coordination: confirm or renegotiate an employee's availability.

All three use the same preview, approval, CALL-E execution, evidence, and human decision flow.

The public demo runs in a deterministic CALL-E sandbox mode. This lets judges test the product without placing a phone call or spending CALL-E credits. The repository also includes an opt-in live adapter for a server-side API key and one authorized E.164 test number.

Where the idea came from

I have a SaaS product in beta that helps businesses handle WhatsApp conversations. That work showed me how much time teams spend on repetitive customer and operations messages. It also gave me a small base of general lessons about state, approvals, and audit trails.

I used only those author-owned, general patterns as a starting point. E-mploye is a new project for this hackathon, not a copy of that SaaS. The phone workflow, CALL-E integration, result handling, safety boundaries, tests, documentation, and demo were built for this submission.

How CALL-E is used

The server uses the official TypeScript SDK, @call-e/calle, through a dedicated provider adapter. In live mode it creates a CALL-E task, reads the call status, and retrieves the structured result, transcript, evidence, and developer events. The API key stays on the server and the manager approval step is required before a provider call.

The public deployment is fake-only by design. The repository includes npm run test:calle, a no-call SDK contract smoke test that verifies the documented create response (201), status response (200), and event response (200) with a local HTTP fixture. This proves the integration contract without calling a real phone.

What I learned

Before this hackathon I did not know CALL-E. I learned how to integrate its TypeScript SDK correctly, keep credentials server-side, use stable idempotency keys, and turn a phone conversation into structured data that a person can review.

I also learned that a phone agent is more useful when it is part of a complete workflow. A call should not silently change a calendar, shift, or CRM record. The manager should see what will happen, what the person said, and then make the final decision.

I found CALL-E especially useful and interesting because it connects an AI agent to real-world work that normally stays in manual phone calls.

Challenges and limitations

The public demo uses fake calls because a public link must be safe and free to test. Live mode is available only when the operator supplies a server-side key, a supported destination region and locale, and a phone number that is owned or explicitly authorized for testing. Carrier delivery and regional availability still need a controlled live test.

This MVP does not claim to be a full CRM, calendar, payroll, or workforce-management product. Those adapters are future work. The important part for this submission is the complete approval-gated phone workflow and the reusable safety model around it.

Thank you

Thank you to the CALL-E team and Devpost for the opportunity to participate. I learned a lot while building this project, and I am excited to keep exploring practical and safe ways to use CALL-E in business workflows.

Built With

• ai
• call-e
• human-in-the-loop
• node.js
• react
• sdk
• typescript
• vite
• voice-ai
• workflow-automation