korayercan/data-science-interview-coach

🧠 Data Science Interview Coach

A Generative AI App for Data Science Interview Practice

🚀 Overview

Data Science Interview Coach is a Retrieval-Augmented Generation (RAG) based application designed to help learners practice and understand data science interview concepts interactively. It uses LangChain, Chroma, and Google Gemini 2.0 Flash to retrieve and generate precise, context-aware answers to user questions about machine learning, statistics, and related topics.

This project was developed as part of the GAIH Generative AI Bootcamp (2025).

🧩 Features

• —🧾 RAG-based Q&A system using your custom dataset of 1,000+ question–answer pairs
• —💬 Conversational Streamlit UI with contextual retrieval
• —🔍 Semantic search powered by bge-small-en-v1.5 embeddings
• —⚙️ Hybrid retrieval strategy (threshold + MMR for diverse results)
• —🌐 Deployed on Hugging Face Spaces for easy access
• —📚 Context sources shown transparently under each answer

📂 Project Architecture

User
 │
 ▼
Streamlit UI  →  Retriever (MMR + threshold)
 │
 ▼
Chroma Vector Store (local persist)
 │
 ▼
Gemini 2.0 Flash Model  →  Response Generation

Key Components:

• —Embeddings: BAAI/bge-small-en-v1.5
• —LLM: Google Gemini 2.0 Flash (via langchain-google-genai)
• —Vector DB: Chroma (persistent local store)
• —Framework: LangChain + Streamlit

🧮 Dataset Preparation

All questions and answers were collected and merged from multiple public sources:

Data cleaning and merging were handled in data_collecting.ipynb:

• —Removed duplicates and empty rows
• —Unified column names (question, answer, source)
• —Added source labels for traceability
• —Final dataset: 1,082 cleaned Q&A entries

Saved as:

data/df_all_qa.csv

⚙️ How It Works

1. 1.User asks a question via Streamlit chat.
2. 2.The app retrieves semantically similar passages using Chroma and MMR retriever.
3. 3.If no relevant context is found, a fallback prompt (Gemini only) generates an answer.
4. 4.The model’s response and source metadata are displayed in the chat interface.

💻 Setup & Run Locally

1️⃣ Clone the repository

git clone https://github.com/<your-username>/data-science-interview-coach.git
cd data-science-interview-coach

2️⃣ Create a virtual environment

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

3️⃣ Install dependencies

pip install -r requirements.txt

4️⃣ Create .env file

GOOGLE_API_KEY=your_google_api_key_here

5️⃣ Run the app

streamlit run App.py

The app will start at http://localhost:8501

🌐 Online Demo

You can try the deployed version directly on Hugging Face Spaces:

👉 [Data Science Interview Coach – Live Demo](https://huggingface.co/spaces/korayercan/data-science-interview-coach)

📦 Repository Structure

Data Science Interview Coach/
├─ App.py                     # Streamlit RAG app
├─ data_collecting.ipynb      # Data cleaning & merging
├─ data/
│   └─ df_all_qa.csv          # Final merged dataset
│   └─ glossary.csv    
│   └─ Mridul-Dixit.csv    
│   └─ team-bay.csv    
│   └─ Yessica.csv    
│   └─ mjphayes.parquet
├─ chroma_genai_qa/           # Local vector store (ignored in Git)
├─ requirements.txt
├─ .env
├─ README.md
└─ .gitignore

📚 Tech Stack

🧠 Example Use Cases

• —AI interview preparation
• —Interactive data science learning assistant
• —Internal corporate Q&A bots with company knowledge base

⚖️ License

MIT License © 2025 – Developed by Koray Ercan This project was created for educational purposes as part of the GAIH Generative AI Bootcamp.

🙌 Acknowledgements

Special thanks to:

• —Global AI Hub (GAIH) mentors for guidance
• —Open-source datasets (Kaggle, Hugging Face)
• —Google for Gemini API access
• —Hugging Face for deployment infrastructure