CallFlow

AI phone-agent SaaS that helps small businesses automatically follow up with new enquiries, qualify leads, and manage conversations through one simple workflow.

Comment

Inspiration

I run Dishari Coaching Centre and have personally experienced how small businesses lose potential customers simply because enquiries are not followed up quickly enough. A parent may enquire about a course and move to another coaching centre within minutes if nobody responds.

That made me think: what if every small business could have its own AI phone agent to follow up with new enquiries?

I built CallFlow around that idea, starting with coaching centres and extending the same workflow to clinics, salons, real estate, and restaurants.

What it does

CallFlow is a universal AI phone-agent platform designed for small businesses.

A business provides its basic information and gets a customized enquiry workflow. When a potential customer submits an enquiry form, CallFlow creates a lead, generates a business-specific conversation prompt, initiates the phone-agent workflow, and tracks the outcome through a dashboard.

The core workflow is:

Business setup → Enquiry form → Lead captured → AI phone workflow → Conversation outcome → Dashboard

CallFlow supports five business types:

  • Coaching centres
  • Clinics
  • Salons
  • Real estate
  • Restaurants

It also includes automatic retry handling for unanswered calls.

The submitted demo includes a mock-success CALL-E path so the complete lead-to-dashboard workflow remains demonstrable when the external CALL-E service is unavailable. The CALL-E integration is isolated behind a dedicated service layer.

How we built it

  • Frontend: Next.js 14, TypeScript, Tailwind CSS, shadcn/ui
  • Backend: Python, FastAPI, asynchronous architecture
  • Database: Supabase PostgreSQL with Row Level Security
  • Authentication: Supabase Auth
  • Phone-agent integration: CALL-E REST API
  • Scheduling: APScheduler
  • AI workflow: Dynamic prompt generation based on business type and business context

The architecture separates the phone-agent integration from the rest of the application, allowing the external phone service to be configured without redesigning the core application.

Challenges we ran into

CALL-E availability

CALL-E was unavailable for part of development. Instead of allowing that external dependency to block development, I isolated the phone integration and created a mock-success path so that the complete application workflow could still be developed and demonstrated.

API integration

Understanding the correct authentication and request structure for the CALL-E REST API required careful testing and debugging.

Dynamic conversations

Different businesses need completely different conversations. A coaching centre should not sound like a restaurant or clinic, so I built a dynamic prompt-generation system that adapts the conversation to the business context.

End-to-end workflow

Connecting enquiry forms, database persistence, phone-agent workflows, webhooks, retry scheduling, and dashboard updates required designing the application as an event-driven workflow rather than simply a frontend application.

Accomplishments that we're proud of

  • Built a complete full-stack AI phone-agent SaaS prototype.
  • Designed a multi-tenant architecture for different small-business categories.
  • Created dynamic business-specific conversation prompts.
  • Built the lead-to-dashboard workflow with webhook support.
  • Added automatic retry handling for unanswered calls.
  • Isolated the CALL-E integration behind a dedicated service layer.
  • Contributed CallFlow to the public CALL-E ecosystem through a reviewed and merged pull request.

What we learned

This project taught us that building an AI phone application is more than connecting an AI API.

We learned how to integrate a phone-agent API with a Python backend, design multi-tenant SaaS architecture, generate contextual prompts dynamically, handle asynchronous workflows and webhooks, and design retry and failure-handling logic.

Most importantly, we learned how to isolate external-service dependencies so that development can continue even when an external API becomes temporarily unavailable.

What's next for CallFlow

  • Validate and enable the production CALL-E integration.
  • Deploy the platform using Vercel and Render.
  • Add WhatsApp-based lead capture.
  • Support additional small-business categories.
  • Add analytics for lead conversion and call outcomes.
  • Add call recordings and transcripts where supported.
  • Improve business-specific conversation customization.
  • Add production monitoring and stronger reliability controls.

Built With

Share this project:

Updates

posted an update —

Live Demo Landing page: https://callflow-delta.vercel.app

Try the AI phone agent yourself: Visit our live public enquiry form for a real business on CallFlow: https://callflow-delta.vercel.app/b/toppers-coaching https://callflow-delta.vercel.app/b/tea Fill in your name, phone number, and a short query — CallFlow's AI agent (powered by CALL-E) will call you within 60 seconds and have a real conversation based on the enquiry.

How CallFlow works end-to-end:

A business signs up at /onboarding — fills 5 fields, gets a custom AI phone agent instantly Their public enquiry page goes live at /b/ Any lead who submits that form gets a real outbound call from CALL-E within 60 seconds Call outcomes, transcripts, and bookings appear on the business's live dashboard

Log in or sign up for Devpost to join the conversation.

Submission history