Retrieval-Augmented Generation (RAG) Q&A App using Streamlit and Wikipedia

WIKI-RAG is a Streamlit-powered Retrieval-Augmented Generation (RAG) app that answers user questions based on Wikipedia articles.
It uses:

• SentenceTransformers to create semantic embeddings.
• FAISS for efficient vector-based retrieval.
• Hugging Face Question-Answering model (roberta-base-squad2) for final answer extraction.

• 🔍 Wikipedia Retrieval – Fetches and chunks Wikipedia content.
• 🧠 QA Model – Answers questions using Hugging Face's roberta-base-squad2.
• 🤖 Semantic Search – Dense embeddings using SentenceTransformers.
• ⚡ Vector Search – FAISS-powered similarity search.

WIKI-RAG/
├── https://github.com/BVNAHUSH/WIKI-RAG/raw/refs/heads/main/advocator/WIK-RAG-v1.3.zip # Streamlit app
├── https://github.com/BVNAHUSH/WIKI-RAG/raw/refs/heads/main/advocator/WIK-RAG-v1.3.zip # Notebook for testing in Colab
├── https://github.com/BVNAHUSH/WIKI-RAG/raw/refs/heads/main/advocator/WIK-RAG-v1.3.zip # Dependencies
├── https://github.com/BVNAHUSH/WIKI-RAG/raw/refs/heads/main/advocator/WIK-RAG-v1.3.zip # Documentation
└── .gitignore

1. Clone the Repository:

   git clone https://github.com/BVNAHUSH/WIKI-RAG/raw/refs/heads/main/advocator/WIK-RAG-v1.3.zip
   cd WIKI-RAG

2.Create Virtual Environment & Install Dependencies:

python -m venv venv
venv\Scripts\activate   # Windows
# or
source venv/bin/activate   # Mac/Linux

pip install -r https://github.com/BVNAHUSH/WIKI-RAG/raw/refs/heads/main/advocator/WIK-RAG-v1.3.zip

3.Run the App

streamlit run https://github.com/BVNAHUSH/WIKI-RAG/raw/refs/heads/main/advocator/WIK-RAG-v1.3.zip

4.Open in Browser:

 http://localhost:8501.

☁ Run in Google Colab You can test the RAG pipeline in Google Colab:

🧪 Example Queries

Topic: Artificial Intelligence

Question: Who is known as the father of AI?

Topic: James Webb Space Telescope

Question: What is the mission of JWST?

Topic: World War II

Question: When did World War II start and who were the Axis powers?

Workflow Steps & Functions:

1. get_wikipedia_content(topic)
   → Fetches Wikipedia article text for the selected topic.

2. split_text(document)
   → Splits the article into smaller overlapping chunks for better retrieval.

3. create_embeddings(chunks)
   → Generates embeddings for all chunks using SentenceTransformer.

4. build_faiss_index(embeddings)
   → Stores embeddings in a FAISS vector index for fast similarity search.

5. retrieve_chunks(query)
   → Encodes the user query and retrieves top k similar chunks from FAISS.

6. answer_question(query, context)
   → Uses Hugging Face QA pipeline (roberta-base-squad2) to find the answer from the retrieved chunks.

7. Streamlit UI
   → Collects user input (topic + question) and displays: - Number of chunks - Retrieved chunks - Final answer