QuickRag

QuickRag - Project Overview

QuickRag is an open-source toolkit for Retrieval-Augmented Generation (RAG).
It combines document indexing, retrieval, optional reranking, and querying large language models (LLMs) like Gemini.

---

🚀 Quick Start

Clone and install

git clone https://github.com/ton_compte/QuickRag.git
cd QuickRag

# optional: create a virtual environment
python -m venv .venv
source .venv/bin/activate  # Linux / Mac
.venv\Scripts\activate     # Windows

# install dependencies
pip install -r requirements.txt
pip install -e .            # editable install for development

---

🏗️ Architecture Overview

The project is structured around four main layers:

---

🚀 Core Components

1. ColbertRAG

Handles document indexing and retrieval with ColBERT.

Key features:

• create_index_from_docs(doc_path, index_name): index documents from PDFs.
• retrieve_docs(query, index_path, top_k): fetch top-k documents.
• query(query, index_path, reranker=None, top_k=5): retrieves documents, optionally reranks, and queries Gemini.

Example usage:

colbert = ColbertRAG()
index_path = colbert.create_index_from_docs("docs/", index_name="my_index")
retrieved_docs = colbert.retrieve_docs("What is neural search?", index_path)

2. Rerankers

Abstract class for document reranking:

from Reranker import Reranker
reranker = Reranker(top_k=5)
ranked_docs = reranker.rank_docs(query="Neural search?", docs=retrieved_docs)

Built-in implementations:

• Reranker: CrossEncoder-based ranking.
• ColbertReranker: Uses ColBERT for reranking.

3. VectorStore

Persistent storage for embeddings using ChromaDB. Supports multiple collections and provides methods to:

• create_collection(name, embedding_model): create or retrieve a collection.

• add_doc_to_collection(docs, collection): add new documents.

• retrieve_docs(query, collection, top_k): retrieve top-k relevant documents.

• delete_collection(name) and list_collection().

db = VectorStore(db_name="my_db")
collection = db.create_collection("my_collection", embedding_model_name="intfloat/multilingual-e5-small")
db.add_doc_to_collection(chunks, collection)
docs = db.retrieve_docs("What is neural search?", collection)

4.QuickRag

High-level wrapper to create a full RAG pipeline:

• Loads and chunks documents.
• Stores embeddings in a vector store.
• Retrieves top-k documents.
• Optionally reranks documents.
• Queries Gemini LLM for answers

Example:

qr = QuickRag()
answer = qr.create_naive_gemini(
    path_documents="docs/",
    query="What is neural search?",
    gemini_model="gemini-2.5-flash",
    reranker=reranker
)
print(answer)