Phone Interview Assistant

An automated phone interview system using Plivo for telephony, Deepgram for speech recognition and AI agent capabilities, and MongoDB for data storage.

Local Testing

1. run: python server.py
2. open any api client like rapid API client
3. paste the /make-a-call endpoint in POST request and pass this body in the same:

{ "phone": 9999999999, "name": "Ayam", "id": "66796c746e6b34001c4480df" }

Deployment Testing

1. paste this endpoint: "https://phoneagent.hyrgpt.com/make-a-call" in postman with POST request
2. use the same sample data as in local testing for the deployment code.

Important Points

1. https://phoneagent.hyrgpt.com is our dev public deployment URL
2. All API Keys are available in the config file
3. DO NOT CHANGE THE PORT ON WHICH THE APP IS RUNNING

System Architecture

Components

• Quart Web Server: Async web framework handling HTTP and WebSocket connections
• Plivo Integration: Handles outbound calls and audio streaming
• Deepgram AI: Provides speech-to-text and AI agent capabilities
• MongoDB: Stores screening data, jobs, questions, and interview transcripts

Database Collections

• phoneScreening: Stores interview transcripts
• screenings: Interview session data
• jobs: Job posting information
• jobquestions: Interview questions for specific jobs
• companies: Company information

Main Features

1. Automated Calling

   • Makes outbound calls to candidates using Plivo
   • Uses configured phone number from config.json

2. Real-time Audio Processing

   • Bidirectional audio streaming
   • 8kHz mu-law audio encoding
   • Buffer management for audio chunks

3. Interview Flow

   • AI agent introduces itself as company representative
   • Conducts interview based on job description and questions
   • Records and transcribes entire conversation
   • Saves interview transcript to database

4. Configuration Requirements

   • config.json file containing:
     • PLIVO_NUMBER
     • DEEPGRAM_API_KEY
     • PLIVO_AUTH_ID
     • PLIVO_AUTH_TOKEN
   • prompt.txt file for AI agent instructions

API Endpoints

1. POST /make-a-call

   • Initiates phone interview
   • Required payload: phone number, candidate name, screening ID
   • Fetches relevant job and company data
   • Starts automated interview process

2. GET/POST /webhook

   • Handles Plivo call events
   • Sets up audio recording and streaming

3. WebSocket /media-stream

   • Manages real-time audio communication
   • Handles bidirectional audio between Plivo and Deepgram

Error Handling

• WebSocket connection management
• Audio stream buffering
• Transcript saving
• Comprehensive error logging

Requirements

• Python 3.9+
• MongoDB
• Plivo account
• Deepgram account
• Required Python packages: plivo, quart, websockets, pydantic, quart-cors

Models;

• Speak: aura-asteria-en
• Think: gpt-4o-mini
• Listen: nova-3