Privacy-Preserving Model Interaction

This repository accompanies our paper:

"PPMI: Privacy-Preserving LLM Interaction with Socratic Chain-of-Thought Reasoning and Homomorphically Encrypted Vector Databases"

While the full interaction system described in the paper will be released later, we have open-sourced the encrypted vector database as the first component. It enables secure vector storage and similarity search over encrypted vectors, forming the foundation of our privacy-preserving framework. Additional components will be released in the future.

Citation

@article{bae2025ppmi,
  title={PPMI: Privacy-Preserving LLM Interaction with Socratic Chain-of-Thought Reasoning and Homomorphically Encrypted Vector Databases},
  author={Bae, Yubeen and Kim, Minchan and Lee, Jaejin and Kim, Sangbum and Kim, Jaehyung and Choi, Yejin and Mireshghallah, Niloofar},
  journal={arXiv preprint arXiv:2506.17336},
  year={2025}
}