🎙️ Interview Learning Assistant

An AI-powered multi-agent system that transforms interview audio recordings into comprehensive learning documents with Q&A pairs and model answers using CrewAI.

📋 Overview

This project implements a sophisticated 6-agent sequential workflow that:

1. Transcribes interview audio using GPT-4o audio preview
2. Extracts every Q&A pair from the interview
3. Generates comprehensive model answers for each question
4. Analyzes performance gaps and learning opportunities
5. Formats everything into a clean study document
6. Exports results as structured markdown files

🚀 Features

• Audio Transcription: Supports MP3, WAV, M4A, OGG, and WebM formats (up to 20MB)
• Complete Q&A Extraction: Captures every single question and answer from the interview
• Model Answer Generation: Provides comprehensive correct answers for all questions
• Topic Organization: Groups related questions into logical sections
• Learning Focused: Designed for interview preparation and knowledge archiving
• Multi-Model Support: Uses optimized AI models for each specific task

🛠️ Installation

Prerequisites

• Python 3.8 or higher
• OpenRouter API key (get one at https://openrouter.ai/keys)

Setup

1. Clone the repository:

git clone https://github.com/yourusername/internship-agent.git
cd internship-agent

2. Create a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. Configure API keys:

cp .env.example .env
# Edit .env and add your OPENROUTER_API_KEY

📖 Usage

Basic Usage

python src/main.py --audio interview.mp3

With Company and Role Information

python src/main.py --audio interview.mp3 --company "Tech Corp" --role "Software Engineer"

Custom Output Directory

python src/main.py --audio interview.wav --output results/

Command Line Options

• --audio: Path to the interview audio file (required)
• --company: Company name for the interview (default: "Unknown Company")
• --role: Role applied for (default: "Unknown Role")
• --output: Output directory for the analysis (default: "interviews/")

🤖 Agent Architecture

1. Transcriptionist

• Role: Audio Intelligence Specialist
• Model: GPT-4o Audio Preview
• Task: Accurately transcribe audio to text

2. Dialogue Structurer

• Role: Interview Q&A Extractor
• Model: Google Gemini 2.5 Flash Lite
• Task: Extract ALL Q&A pairs and organize by topic

3. Technical Architect

• Role: Model Answer Creator
• Model: Google Gemini 2.5 Flash Lite
• Task: Generate comprehensive model answers for every question

4. Performance Analyst

• Role: Learning Performance Analyst
• Model: Google Gemini 2.5 Flash Lite
• Task: Compare answers and create learning roadmap

5. Quality Editor

• Role: Learning Document Creator
• Model: Google Gemini 2.5 Flash Lite
• Task: Format Q&As with model answers clearly

6. Notion Exporter

• Role: Digital Records Manager
• Model: Google Gemini 2.5 Flash Lite
• Task: Export and verify completeness

📂 Project Structure

internship-agent/
├── src/
│   ├── agents/         # Agent definitions
│   ├── tasks/          # Task configurations
│   ├── tools/          # Custom tools (transcription, file writer, Notion)
│   ├── crew.py         # Main crew orchestration
│   └── main.py         # CLI entry point
├── config/
│   ├── agents.yaml     # Agent configurations
│   └── tasks.yaml      # Task configurations
├── audio/              # Input audio files
├── interviews/         # Output analysis reports
├── requirements.txt    # Python dependencies
├── .env.example        # Environment variables template
└── README.md          # This file

📊 Output Format

The learning document includes:

# Interview Learning Document

## Interview Metadata
- Company: [Company Name]
- Role: [Position]
- Interviewer: [Name]
- Candidate: [Name]

## Topic: [e.g., Options Pricing]

### Question 1: [Exact question text]

**Candidate's Answer:**
[Complete answer from candidate]

**MODEL ANSWER:**
[Comprehensive correct answer with explanations]

**Key Learning Points:**
- Concept 1
- Concept 2

---

🔧 Configuration

Environment Variables (.env)

# Required
OPENROUTER_API_KEY=your_openrouter_api_key_here
OPENROUTER_BASE_URL=https://openrouter.ai/api/v1

# Optional (for Notion export)
NOTION_TOKEN=your_notion_token_here
NOTION_DATABASE_ID=your_notion_database_id_here

Model Configuration

Models can be changed in config/agents.yaml. Current models used:

• Transcription: openai/gpt-4o-audio-preview
• All Agents: google/gemini-2.5-flash-lite (cost-effective and efficient)

🧪 Testing

To test with a sample audio file:

1. Place an interview audio file in the audio/ directory
2. Run: python src/main.py --audio audio/sample_interview.mp3
3. Check the interviews/ directory for the output

🤝 Contributing

This is a homework project, but suggestions and improvements are welcome!

📄 License

MIT License - See LICENSE file for details

🔮 Future Enhancements

• Coding interview support
  • After transcription, new agent figures out is it live-coding interview or not
  • If it is, it routes to another agent that solves coding tasks
  • After solving, it routes to another agent that checks the solution inside a code sandbox

🙏 Acknowledgments

• CrewAI framework for multi-agent orchestration
• OpenRouter for unified AI model access
• MIT AI Studio for the assignment framework

---

Built with CrewAI | Powered by OpenRouter