Smart Interview

An AI-powered mock interview app that reads your resume and asks you personalized technical questions based on your actual projects, skills, and experience — just like a real recruiter would.

How it works

1. Upload your resume (PDF)
2. The app parses it into sections (Skills, Projects, Experience, Education, etc.)
3. Chunks are embedded and stored in an in-memory vector database (ChromaDB)
4. A Groq-powered LLM generates 8 targeted interview questions referencing your specific work
5. For each question, it generates a follow-up based on your answer
6. Questions vary between sessions — different interviewer personas and shuffled context mean no two sessions are identical

Stack

Layer	Tool
UI	Streamlit
LLM	Groq (llama-3.3-70b-versatile)
Vector DB	ChromaDB (in-memory)
PDF Parsing	pdfplumber
Embeddings	ChromaDB default (MiniLM)

Setup

1. Clone the repo

git clone https://github.com/FaresIbrahim32/smart_interview.git
cd smart_interview

2. Create and activate a virtual environment

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

3. Install dependencies

pip install -r requirements.txt

4. Add your Groq API key

Create a .env file in the project root:

RAG_API=your_groq_api_key_here

Get a free API key at console.groq.com.

5. Run the app

streamlit run app.py

Features

• Section-aware chunking — resume is split by detected sections so questions are targeted
• Randomized personas — each session picks a different interviewer style (startup CTO, staff engineer, etc.)
• Follow-up questions — after each answer, a contextual follow-up is generated
• Progress tracking — sidebar shows all questions with completion status
• No data stored — everything is in-memory, cleared when you close the session

Project structure

smart_interview/
├── app.py               # Streamlit UI and interview loop
├── requirements.txt
├── rag/
│   ├── parser.py        # PDF extraction and section-based chunking
│   ├── vectorstore.py   # ChromaDB setup and querying
│   └── interviewer.py   # Groq LLM calls for question and follow-up generation