inha-greedy

FHE-RAG: Fully Homomorphic Encryption for Retrieval Augmented Generation

Welcome to the FHE-RAG project, a research initiative by students from the Department of Information and Communication Engineering at Inha University.
Our goal is to explore the integration of Fully Homomorphic Encryption (FHE) with the Retrieval Augmented Generation (RAG) framework, enabling secure and privacy-preserving language generation.

Retrieval Augmented Generation (RAG) is a powerful technique that combines information retrieval and language generation models to produce accurate and informative outputs.
However, the document corpus used in RAG may contain sensitive data, raising privacy concerns.

Our project aims to address this issue by applying Fully Homomorphic Encryption (FHE) to the RAG pipeline.
FHE allows arbitrary computations to be performed on encrypted data without the need for decryption, ensuring end-to-end data confidentiality.

Encrypted Document Corpus: The document corpus used for retrieval is encrypted using a Fully Homomorphic Encryption scheme, protecting sensitive information.
Secure Retrieval and Encoding: The retrieval and encoding processes are performed entirely on encrypted data, maintaining data privacy throughout the pipeline.
Compatibility with RAG Frameworks: Our implementation is designed to seamlessly integrate with existing RAG frameworks, minimizing the need for significant architectural changes.