Pass Interview

Demo of an LLM agent passing a technical interview. Uses OpenAI Realtime API for voice and a Chrome extension for typing.

Backend forked from https://github.com/openai/openai-realtime-console.

Prerequisites

• OpenAI API key (you will be prompted to enter it when app is started)
• Browser extension in Google Chrome (enter path to it: browser_extension/)
• *(For MacOS) Blackhole-16ch to bridge audio from Google Meet to React app and back

*Support for other platforms will be added soon

Installation

1. Clone the repository:

   git clone https://github.com/PalisadeResearch/pass-interview.git
   cd pass-interview

2. Install dependencies:

   npm i

3. Create two new virtual audio devices:

   • Open Audio MIDI Setup
   • Create two aggregate devices
   • In each of them label channels either 1 and 2 or 3 and 4
   • Name them accordingly
   • Set first device as mic in Google Meet
   • Set second device as default system speaker in Settings

4. Configure the Chrome Extension:

   • Install the extension from browser_extension/
   • Copy your extension ID from Chrome (chrome://extensions/)
   • Update the extension ID in src/pages/ConsolePage.tsx:

   chrome.runtime.sendMessage(
     'YOUR_EXTENSION_ID_HERE', // Replace with your extension ID
     { type: 'CODE_FROM_REACT', code: code },

5. Start the application:

   • Create a new Repl at replit.com
   • Run the development server:

   npm start

   • Enter your OpenAI API key when prompted

Customization

System Instructions

You can modify the AI agent's behavior by editing the instructions in src/utils/conversation_config.js. This file contains the system instructions that define how the agent should behave and respond.

Code Display Animation

The code typing animation can be customized in src/pages/ConsolePage.tsx by modifying the typeCode function.

Initial Conversation Message

You can modify the first message sent to the AI agent by editing the connectConversation function in src/pages/ConsolePage.tsx.