ChainCloak

Privacy and LLMs! But just one centralized layer is not enough, we imagine a future that anyone can deploy their own Anonymization node, share that in a contract, and users freely choose among them.

Comment

Inspiration

Our core idea revolves around the concept of providing private layers for Large Language Models (LLMs). We believe that privacy is essential in today's data-driven world, and centralized solutions are not sufficient. Our inspiration stems from envisioning a future where anyone can deploy their own Anonymization node, share it in a smart contract, and give users the freedom to choose among them.

What it does

Our project demonstrates the power of decentralized Anonymization nodes for LLMs. We have deployed different layers using OpenAI and Cohere, one focusing on privacy and the other not. Through our front-end interface, we showcase how user experiences can vary based on their choice of Anonymization module.

In the future, we envision these nodes to be highly customizable, allowing each Anonymization node to incorporate Natural Language Processing (NLP) modules for extracting sensitive inputs from prompts, making the process even more secure and user-friendly.

How we built it

Our project is built on a decentralized architecture. Here's a high-level overview of how it works:

  1. User Interaction: Users input their queries into an LLM-enabled device.

  2. Deploy Anonymization Node: The query is sent to a Custom node (based on their reputation), where identifiers and sensitive information are further anonymized.

  3. LLM Processing: The anonymized query is forwarded to the LLM provider for processing.

  4. Data Enrichment (In Future): The LLM provider sends the response back to the custom node. The node then injects the sensitive information back into the response.

  5. User Experience: The enriched response is sent back to the user's device, ensuring privacy and a seamless user experience.

Challenges we ran into

While building our decentralized Anonymization system, we faced various technical challenges, including:

  1. Figuring out a way to use deployed smart contract as a registry for available nodes.

  2. Connecting all three components (backend, frontend, private layer) in a manner that does not hurt user experience.

Accomplishments that we're proud of

  • Successfully deploying decentralized Anonymization nodes.
  • Demonstrating how user experiences can be enhanced with privacy-focused solutions.
  • Designing a system that can adapt and evolve with future NLP modules.

What we learned

Throughout this project, we gained valuable insights into decentralized systems, smart contracts, and the importance of user privacy. We also learned how to work with APIs provided by two LLM giants (cohere and openai)

What's next for ChainCloak

The future of ChainCloak looks promising. We plan to:

  • Expand the range of Anonymization modules and LLM providers.
  • Enhance the security and customization options for Anonymization nodes.
  • Collaborate with the community to build a robust ecosystem of privacy-focused solutions.
  • Continue exploring new technologies and innovations in the field of decentralized AI and privacy.

We are excited about the potential impact of ChainCloak in ensuring privacy in the era of AI-powered language models.

Share this project:

Updates