Lord of the solvers
FHE and ML powered user intent solver, ensuring user privacy and facilitating web2 like experience.
Project Description
Introducing an innovative tool powered by Fully Homomorphic Encryption (FHE) designed to revolutionize the way we solve intents. At its core, this tool understands user intents, deciphering the desired outcomes, whether it's swapping tokens or acquiring specific NFTs.
Solvers, the backbone of this system, efficiently pick transactions aligned with user intents. Their mission is clear: to help users save valuable time and steer clear of potential frustrations. Imagine a seamless experience reminiscent of Web2, made even better with the help of account abstraction, where solving intents becomes a breeze.
However, traditional solvers pose challenges, particularly in terms of user privacy. Intentions are laid bare, exposing users to potential risks such as frontrunning and MEV attacks. Our FHE-powered solution addresses these concerns, ensuring a secure and private environment for users to navigate their intentions without compromising confidentiality. Welcome to the future of intent-solving, where privacy and efficiency coexist harmoniously.
How it's Made
The process begins with the creation of a comprehensive dataset comprising sample swap and send intents, laying the foundation for a machine learning model. Leveraging a tf-idf vector space retrieval-based approach, the model is meticulously trained to accurately classify intents into their respective categories.
The heart of this innovative solution lies in the integration of Zama's Concrete ML, specifically utilizing the XGBoost classification algorithm. The trained model, now equipped with the ability to precisely categorize intents, serves as the intelligent solver in the subsequent steps.
Upon user input of intents, a unique set of Fully Homomorphic Encryption (FHE) keys is generated for each session, ensuring a secure and private environment. The intent is then encrypted and transmitted to the FHE solver. This solver, powered by the Concrete ML model, processes the encrypted intent using the XGBoost classification, providing an encrypted solution.
The encrypted solution, along with the client model, is transmitted back to the user. Upon receiving the encrypted package, the client undertakes the decryption process, revealing the clear solution. This decrypted solution is then transformed into a transaction for on-chain settlement.
What sets this process apart is the commitment to privacy and user experience. The entire settlement occurs gas-free, facilitated by Pimlico bundlers and paymasters. This is how intent reaches settlement with privacy and good UX!
