Proof of Concept

To develop the proof of concept (PoC), we are implementing an alternative architecture designed to demonstrate the core functionality of our decentralized vector storage system. Key Workflow:

  1. User authentication and interaction Implement a secure login system with funds locking to prevent sybil attacks and dataset poisoning. Users can interact with the service via API or a React-based front end with templates and UI kits. The front end includes a dashboard for tracking transactions, assessing system status, and handling errors.

  2. Request processing User requests in natural language are sent to an embedding model for vectorization. Embedding models are deployed on three servers to ensure decentralization and achieve consensus through random node selection and timeouts.

  3. Embedding and search Utilize the BERT model for embedding, fine-tuned for consistent performance. Deploy Milvus or qDrant as search nodes on three servers, using ANN search algorithms, random selection, and node timeout for embedding creation. Vector data is stored in Milvus or Chroma DB across three servers, with configured collections and search distances.

  4. Reward distribution Results from vector searches are sent to a smart contract, which distributes rewards to miners based on their participation. Funds for rewards are deducted from the requester's wallet according to a predefined formula. Every vector search triggers a transaction on the smart contract without validation, ensuring seamless reward distribution.

This PoC will result in a fully decentralized, secure, and scalable vector storage system. Users can deploy nodes and earn rewards for creating embeddings, performing vector searches, and storing data. Additionally, we will develop a dashboard for transaction visualization, integrated with an L2 solution.

Last updated