Quantum-RAG

A Python semantic-retrieval prototype with separate quantum-inspired experimental modules.

This project explores the retrieval stage of retrieval-augmented generation: split documents into chunks, embed them, search for relevant passages, and assemble context for an LLM.

Architecture

Document → chunking → embeddings → vector store
                                      ↑
Query → query embedding → ranked passages → LLM context

The public RAG entry point connects chunker.py, embedder.py, and vector_store.py. Generation is separate from retrieval.

Quick start

git clone https://github.com/asadsehto/Quantum-RAG.git
cd Quantum-RAG
python -m venv .venv

Activate your virtual environment, then install:

python -m pip install -r requirements.txt
python -m pip install -e .

from quantum_rag import RAG

rag = RAG()
rag.add_document("Semantic retrieval uses embeddings to find relevant passages.")

for result in rag.search("How are relevant passages found?", top_k=3):
    print(result["score"], result["text"])

context = rag.get_context("How are relevant passages found?")

Embedding-model downloads depend on the configured implementation and runtime environment.

Repository map

Path	Role
quantum_rag/chunker.py	Document chunking
quantum_rag/embedder.py	Text embeddings
quantum_rag/vector_store.py	Vector storage and search
quantum_rag/rag.py	Indexing, retrieval, and context assembly
quantum_rag/quantum/	Experimental complex embeddings, similarity, density, and fusion
examples/	Usage examples

Experimental scope

The quantum-inspired components are separate experiments. The default RAG entry point uses conventional embedding and vector-store modules; it does not itself wire in those experimental components. This is quantum-inspired software, with no quantum hardware execution claimed.

Evaluation

No verified improvement over a classical retrieval baseline is claimed. A reproducible comparison should hold the dataset, embedding model, chunking, and retrieval depth constant, then measure Recall@k, ranking quality, latency, and memory for each explicit retrieval path.

Author

Asad Saleem · LinkedIn