🤖📚 AI Study Assistant - RAG Chatbot with FastAPI & Gemini AI

A smart document-based study assistant built with Local Embeddings, Vector Search (FAISS), and Gemini AI. This app allows users to upload documents (PDFs, DOCX, PPTX, images), extract content using OCR, embed it locally, store it in a vector store, and ask natural language questions. Features a hybrid architecture combining cloud AI intelligence with local processing for privacy and speed.

A practical implementation of Retrieval-Augmented Generation (RAG) with a privacy-first approach.

Architecture

Core Workflow

1. Upload Documents (PDF, DOCX, PPTX, images via OCR)

2. Extract & Process Text locally using various parsers and Tesseract OCR

3. Chunk & Embed using Local SentenceTransformers (all-MiniLM-L6-v2)

4. Store in FAISS vector database for semantic search

5. Receive Query → Search vector store for relevant context

6. Generate Answer using Gemini AI with retrieved context

7. Create Study Materials (MCQs, flashcards, summaries)

📁 Project Structure

ai-study-assistant/
│
├── 📄 streamlit_app.py              # Streamlit frontend interface
│
├── 📁 backend/                      # FastAPI backend
│   ├── main.py                      # FastAPI application & routes
│   ├── config.py                    # Configuration settings
│   ├── faiss_store.py               # FAISS vector database operations
│   ├── embeddings.py                # Local embedding generation (SentenceTransformers)
│   ├── qa_engine.py                 # Q&A with Gemini AI & context retrieval
│   ├── mcq_generator.py             # MCQ generation using Gemini
│   ├── flashcard_maker.py           # Flashcard creation
│   ├── summarizer.py                # Text summarization
│   ├── file_extractor.py            # Document text extraction (PDF, DOCX, PPTX, OCR)
│   ├── rag_pipeline.py              # RAG processing pipeline
│            
│
├── 📁 vectorstore/                  # FAISS vector store (auto-created)
│   └── faiss_index/                 # Index and metadata storage
│
├── 📁 uploaded_docs/                # Uploaded documents storage (auto-created)
├── 📄 .env                          # Environment variables
├── 📄 .gitignore                    # Git ignore file
├──📄 README.md                 
└── requirements.txt

Setup Instructions

1. Clone & Navigate

git clone https://github.com/m-sameerkhan/ai-study-assistant.git
/ai-study-assistant.git
cd ai-study-assistant

2. Create Virtual Environment & Install Dependencies

# Windows
python -m venv venv
venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

3. Install Required Tools

Tesseract OCR (for image text extraction)

• Download from: https://github.com/UB-Mannheim/tesseract/wiki

• Add to system PATH

4. Configure Environment Variables

Create a .env file in the project root:

# Gemini API Key (Get from: https://ai.google.dev/)
GEMINI_API_KEY=your_gemini_api_key_here

# Gemini Model (Use gemini-pro for free tier)
GEMINI_CHAT_MODEL=gemini-pro

# Local Embeddings Model
EMBEDDING_MODEL_NAME=all-MiniLM-L6-v2

# Storage Directories
FAISS_DIR=./vectorstore/faiss_index
UPLOAD_DIR=./uploaded_docs

# Text Processing
CHUNK_SIZE=800
CHUNK_OVERLAP=150
MAX_RESPONSE_TOKENS=500
TEMPERATURE=0.2

# Server Configuration
HOST=0.0.0.0
PORT=8000
DEBUG=True
API_BASE=http://127.0.0.1:8000

5. Run the Backend (FastAPI)

uvicorn backend.main:app --host 0.0.0.0 --port 8000 --reload

6. Run the Frontend (Streamlit)

# Open new terminal
streamlit run streamlit_app.py

How It Works

1. Document Upload & Processing

• Upload PDF, DOCX, PPTX, TXT, or images (PNG, JPG)

• PDFs/DOCX/PPTX: Extracted using specialized libraries

• Images: Processed with Tesseract OCR

• Text is cleaned, chunked, and prepared for embedding

2.Local Embedding & Storage

• Each text chunk is converted to embeddings using all-MiniLM-L6-v2

• Embeddings are stored in FAISS vector database locally

• Metadata preserves document references and chunk relationships

3. Query Processing

• User question is embedded using the same local model

• FAISS performs similarity search to find relevant document chunks

• Top-k most relevant chunks are retrieved as context

4. Intelligent Response Generation

• Retrieved context + user question sent to Gemini AI

• Gemini generates accurate, context-aware responses

• Confidence scores indicate answer reliability

5. Study Material Creation

• MCQs: Generate multiple-choice questions with explanations

• Flashcards: Create Q/A pairs for active recall practice

• Summaries: Produce concise summaries in various formats

Example Usage

1. Upload Study Materials

• Upload your textbook PDFs, lecture notes, or research papers
• System extracts text and creates searchable embeddings

2. Ask Questions

Q: "What are the key differences between supervised and unsupervised learning?"
A: [Based on your uploaded machine learning textbook] 
   Supervised learning uses labeled data with clear input-output pairs...

3. Generate Practice Materials

• MCQs: "Generate 5 questions about neural networks"
• Flashcards: "Create flashcards for Actuators chapter 2"
• Summaries: "Summarize this 10-page article in bullet points"

Key Dependencies

• fastapi, uvicorn (Web framework & server)

• sentence-transformers (Local embedding models)

• faiss-cpu (Vector similarity search)

• google-generativeai (Gemini AI integration)

• pypdf2, python-docx, python-pptx (Document parsing)

• pytesseract, pillow , opencv-python (Image OCR processing)

• python-multipart (File upload handling)

• streamlit (Web interface)

• requests (API communication)

Troubleshooting

Issue	Solution
ModuleNotFoundError: No module named backend	Run commands from the project root directory
Permission denied saving FAISS index	Run terminal as Administrator or fix folder permissions
Gemini API errors	Check API key validity and quota in Google AI Studio
Tesseract OCR not working	Ensure Tesseract is installed and added to PATH
Slow first run	SentenceTransformers downloads model (~90MB) on first use
Port already in use	Change port in .env or run with --port 8001

---

Author

Muhammad Sameer Khan
LinkedIn

Contributing

Contributions and forks are welcome!
If you'd like to extend this chatbot with:

• Additional document format support
• More export options (Anki, PDF, etc.)
• Mobile-responsive UI improvements
• Docker deployment scripts
• Performance optimizations
• Additional local LLM options

Feel free to fork or open a pull request.