Proofs of Inference

Problem Statement

This project is a decentralized marketplace for Zero-Knowledge Machine Learning (zkML) proofs, enabling users to request, verify, and pay for cryptographically secure, privacy-preserving model inferences. Model providers quantize and compile their machine learning models into zk-SNARK circuits, generating proving and verifying keys. When a user requests a proof, they stake payment in a smart contract escrow and submit a hashed input. The backend re-executes the inference, generates a zk-proof, and stores it along with audit logs on decentralized storage like Filecoin or Akave. Users can then verify the proof either locally or on-chain, and the smart contract releases payment to the provider only if the proof is valid—otherwise, the user is refunded. The frontend allows users to inspect proof history, request new proofs, and verify results. This architecture ensures a transparent, trustless, and auditable workflow for both users and model providers, leveraging Solidity, Filecoin/Akave, and zkML tooling for robust on-chain/off-chain coordination.

Solution

This project integrates several cutting-edge technologies to create a decentralized zkML proof marketplace. We use ezkl to quantize and compile machine learning models into zk-SNARK circuits, enabling privacy-preserving inference proofs. The backend, built with Node.js and Python, handles proof requests, inference execution, and proof generation, then uploads proofs and logs to decentralized storage solutions like Filecoin and Akave. Solidity smart contracts manage payment escrow and proof verification on-chain, ensuring trustless settlement between users and model providers. The frontend is built in React, providing an interface for proof requests, history, and local/on-chain verification. Our architecture leverages decentralized persistence and cryptographic guarantees, allowing transparent, auditable, and censorship-resistant AI inference as-a-service.