ProofPal

Proofpal is a decentralized zkML (zero-knowledge machine learning) platform designed to bring verifiability, privacy, and incentives to the world of AI inference on blockchain.

In the current landscape, AI usage lacks transparency and trust. When users interact with AI models hosted on centralized platforms like HuggingFace, they are forced to blindly trust the provider. There is no guarantee that the inference was generated by the intended model, no cryptographic verification of results, and no system of incentivization that rewards model creators fairly. Additionally, AI inference on blockchain is still immature, as verifiability is costly and privacy is non-existent. These challenges make it harder to adopt AI in decentralized environments.

Proofpal addresses these problems by embedding zero-knowledge proofs into the AI inference process. When a user queries a model through Proofpal, the system generates not only the inference result but also a zkProof that can be verified on-chain. This ensures that the output indeed comes from the claimed model without exposing sensitive data or requiring blind trust. By leveraging Arbitrum Layer 2, Proofpal achieves low-cost on-chain verification, making zkML both scalable and practical.

The workflow is straightforward:

• a customer chooses a model from a provider, pays for usage through a smart contract, and receives both the inference result and its zkProof.

• model providers are incentivized since they can fix rates for their models and receive payments transparently. Users, on the other hand, gain confidence in correctness, thanks to verifiable proofs on-chain.

To showcase the concept, a demo model has been developed — a basic health risk score predictor. This model, trained in PyTorch with dummy data, takes three inputs (cholesterol, sugar, and blood pressure levels) and outputs a health score. The model is compiled into ONNX circuits, which then produce zkProofs for inference. This MVP demonstrates how Proofpal can blend AI with cryptographic verifiability

Looking ahead, Proofpal aims to integrate Fully Homomorphic Encryption (FHE) so that users can query models with fully encrypted data, preserving privacy end-to-end. It also plans to incentivize decentralized compute providers who perform inference. In contrast to existing centralized AI hubs, Proofpal champions “Trust the Math, not the Platform,” making it a pioneering step toward decentralized, verifiable, and privacy-preserving AI.

depolyed contracts on arbitrum sepolia:

Halo2VerfierCircuit.sol - 0xf8c55Cc643280a486312cd2c8785f47d1c793e84

ModelRegistry.sol- 0x3AF0380edA403c7563f5699178cEFE930907C2De

github link - https://github.com/dumprahul/proofpal

demo video -