A project to get a live transcription of a Twilio phone number using OpenAI Realtime API. Built for BridgeGood 2025 AI for social good hackathon to address scam calls to seniors.
- Vite for frontend
- Backend is FastAPI and websockets for real-time transcription
- Twilio for phone number
Fill out the fields in .env.example and rename to .env.
Then, start ngrok in a terminal tab:
ngrok http 8000
In another terminal tab, cd guardian_line and start the backend server with
uvicorn server:app --host 0.0.0.0 --port 8000 --reload
Finally, cd frontend and then
npm -i to install packages and then
npm run dev to start the frontend.
When you call the twilio number and click on the notification, you should see a live transcript highlighting keywords that can signal a scam. Afterwards, you can see a report.
Here is a loom demonstrating the flow: https://www.loom.com/share/f909d255931c41aeb77228fc34e65b9c