WikiQA: A RAG Pipeline Powered by Google Search and Wikipedia

Overview

WikiQA is a Retrieval-Augmented Generation (RAG) pipeline built for answering questions by retrieving the most relevant wikipedia articles using the Wikipedia API and the Google Search API.

---

Libraries

• Gemini: Used for generating an eval dataset and as a judge.
• LlamaIndex: Core framework for building the RAG pipeline.
• HuggingFace: Open-Source Phi-3.5 SLM and bge-base-en embeddings.
• ChromaDB: Open-source persistent vector DB.
• Ragas: Evaluating the RAG pipeline.

Installation

1. Clone the repository:

   git clone https://github.com/ahmedo42/wikiqa.git
   cd wikiqa

2. Install dependencies:

   pip install -r requirements.txt

3. Run the pipeline demo:

   jupyter notebook demo.ipynb

File Structure

• rag.py: Core RAG pipeline implementation.
• helpers.py: Utility functions for data preprocessing and query handling.
• build_eval_dataset.ipynb: Notebook to construct a dataset for testing using Google's Gemini.
• demo.ipynb: Interactive demo of the pipeline.
• eval.ipynb: Evaluate the RAG pipeline using the generated eval dataset.