- 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.
