Apppublic

kazakiakayami/ai-document-assistant

sourceHugging Faceupdated 7mo agoView on Hugging Face

0likes

App README

📄 AI Document Assistant

An intelligent document assistant powered by RAG (Retrieval-Augmented Generation) that allows you to upload PDF documents and ask questions about their content in natural language.

🚀 Live Demo

🤗 Try it on HuggingFace Spaces

✨ Features

—📁 Upload any PDF document
—💬 Chat with your document in natural language
—🔍 Accurate answers based strictly on document content
—📄 Source page references for every answer
—🌐 Supports both English and Indonesian questions
—🔄 Switch documents anytime without refreshing

🛠️ Tech Stack

Component	Technology
Framework	LangChain
LLM	Groq (Llama 3.1 8B) - Free
Embedding	HuggingFace `all-MiniLM-L6-v2`
Vector DB	ChromaDB
UI	Streamlit
PDF Parsing	PyPDF

🔄 RAG Pipeline

📄 PDF Upload
     ↓
📥 Document Loader (PyPDF)
     ↓
✂️ Text Chunking (chunk_size=1000, overlap=100)
     ↓
🔢 Embedding (HuggingFace sentence-transformers)
     ↓
🗄️ Vector Store (ChromaDB)
     ↓
❓ User Query → Similarity Search → Top 3 Chunks
     ↓
🤖 LLM (Groq/Llama3.1) → Answer + Sources

📁 Project Structure

ai-document-assistant/
│
├── app.py                  # Main Streamlit UI
├── requirements.txt        # Dependencies
├── .env                    # API Keys (not committed)
├── .gitignore
│
└── src/
    ├── document_loader.py  # PDF loading
    ├── chunker.py          # Text chunking
    ├── embedder.py         # Embedding + VectorDB
    └── rag_chain.py        # RAG pipeline + LLM

⚙️ Installation & Setup

1. Clone the repository

bash

git clone https://github.com/YOUR_USERNAME/ai-document-assistant.git
cd ai-document-assistant

2. Create virtual environment

bash

conda create -n rag_env python=3.11
conda activate rag_env

3. Install dependencies

bash

pip install -r requirements.txt

4. Setup environment variables

Create a .env file in the root directory:

GROQ_API_KEY=your_groq_api_key_here

Get your free Groq API key at groq.com

5. Run the app

bash

streamlit run app.py

🧪 Testing Results

Test Type	Question	Result
✅ Positive	"What is the offside rule?"	Accurate answer with sources
✅ Positive	"How many players in a football team?"	"Maximum 11 players"
✅ Negative	"What is the NBA rules?"	"Cannot find answer in document"
✅ Edge Case	"Tell me everything about football"	General summary from context
✅ Boundary	"List all offences in football"	Partial list with honest disclaimer

🔑 Key Design Decisions

—Chunk size 1000, overlap 100 → optimal for legal/rule documents
—k=3 retrieval → balance between context and token efficiency
—Temperature 0.2 → low temperature for factual accuracy
—In-memory VectorDB → no persistence needed for demo deployment
—Prompt engineering → strict instruction to not hallucinate beyond context

📝 Lessons Learned

—RAG accuracy heavily depends on chunk size and overlap configuration
—PDF parsing can introduce artifacts (headers, footers) that affect chunk quality
—Filtering chunks shorter than 50 characters significantly improves retrieval quality
—LLM temperature should be low (0.1-0.3) for document Q&A tasks

👨‍💻 Author

Ahmad Mustofa Z Hacktiv8 Data Science Bootcamp - Phase 1 Project