A conversational AI agent built for the Eightfold Agentic AI Internship assignment.
This project allows users to practice interviews with a realistic AI interviewer and receive detailed, professional feedback.

• Conducts mock interviews for selected job roles
• Asks contextual follow-up questions
• Adapts to user response style (efficient, confused, chatty)
• Provides detailed structured feedback (strengths, areas to improve, tips, rating)
• Voice support (optional – but included in frontend design)
• Fast, free LLM using Groq API (Llama-3 models)

• FastAPI
• Groq API (llama-3.1-8b-instant model – FREE)
• Pydantic
• Uvicorn
• dotenv

• HTML
• CSS
• JavaScript
• Browser Speech-to-Text (Web Speech API)
• Live Server

interview-practice-agent/
│
├── backend/
│   ├── main.py
│   ├── requirements.txt
│   ├── .env
│   └── venv/
│
└── frontend/
    ├── index.html
    ├── style.css
    └── script.js

git clone <your-repo-url>
cd interview-practice-agent

cd backend
python -m venv venv
venv\Scripts\activate

pip install -r requirements.txt

Create a .env file inside backend/:

GROQ_API_KEY=your_key_here

Get your free API key: https://console.groq.com/keys
(No credit card required)

uvicorn main:app --reload

Backend will run at:

👉 http://127.0.0.1:8000

Check health:

👉 http://127.0.0.1:8000/health

cd ../frontend

Right-click index.html →
👉 Open With Live Server

Or open manually:

👉 http://127.0.0.1:5500/frontend/index.html

1. Select job role, experience level, and mode
2. Click Start Interview
3. Answer questions (text or voice)
4. AI interviewer asks smart follow-ups
5. Click Get Feedback
6. Receive:
   • Strengths
   • Areas to improve
   • Actionable tips
   • Rating (0–5)

Frontend (HTML/JS) 
     ↓ REST API calls (fetch)
Backend (FastAPI)
     ↓
Groq LLM (Llama 3.1 Models)

• Frontend sends /start, /message, /finish
• Backend stores session → generates next question
• Feedback is created using structured prompt engineering

• Fastest free inference
• No quota issues
• Great for conversational tasks

• Lightweight
• Easy routing
• Async for smooth processing

• No external dependency
• Works in browser
• Simple to integrate

Used in-memory dictionary:

SESSIONS = { sessionId: { role, level, mode, qa[] } }

Your demo video (max 10 minutes) should show:

• Efficient user
• Chatty user
• Confused user
• Off-topic user

No slides needed — only screen recording with voice.

• Conversational quality
• Agentic behaviour
• Multi-persona handling
• Technical implementation
• Natural interaction
• Detailed README
• Clear architecture reasoning

Modi Sireesha
Eightfold Agentic AI Internship – Assignment Submission