Skip to project
Nishant JagannathGet in touch
← All projects

[ AI / Cryptographic verification ]

Verifiable AI

Can an AI model support its claims without revealing its secrets?

Read the research preprint (opens in a new tab)
My role
First author and researcher
Status
Research preprint
Year
2025
Methods & tools
Zero-knowledge proofszk-SNARKsChainlinkBlockchain

A SHORT RESEARCH FILM

A claim. A proof. A record.

00:20 / SILENT

Automatic motion is off. Play the film when ready.

A short visual journey through private model claims, zero-knowledge proofs, oracle-sourced context and an on-chain research demonstration.

Research preprint · illustrative film
Read the film transcript

  1. An AI model. A claim.

    A model-related claim becomes something to examine. A glowing claim card appears beside a closed model vault.

  2. Private stays private.

    The model vault remains closed. The research investigates verification while protecting proprietary model information.

  3. Turn a claim into proof.

    Particles gather into a cryptographic proof. Zero-knowledge evidence addresses a specific statement; it does not certify every property of an AI system.

  4. Verify with context.

    The proof and claim reach a verifier. An oracle provides externally sourced context within the proposed workflow.

  5. Proof on record.

    A verification record joins a testnet ledger. The 2025 preprint demonstrates a linear-regression example on Sepolia; this is not production deployment or general-purpose AI certification.

Download film

[ 01 / The context ]

The question.
The approach.

AI marketplaces need ways to examine model claims while protecting proprietary information. This research investigates on-chain verification with zero-knowledge proofs and decentralised oracles, connecting cryptographic evidence with externally sourced data.

How the work comes together

Enhancing Trust in AI Marketplaces: Evaluating On-Chain Verification of Personalized AI models using zk-SNARKs was released as a first-author arXiv preprint in 2025. This work connects to my doctoral research on AI verification on blockchain through zero-knowledge proofs and oracles.

A claim to check

Explore how claims about a personalised AI model can be represented in a verifiable workflow.

Privacy by construction

Use zero-knowledge proofs to investigate verification without exposing private model information.

An on-chain demonstration

Combine zk-SNARKs and Chainlink oracles in a linear-regression demonstration on the Sepolia testnet.

[ 02 / Research scope ]

Where the work stands.

A research preprint and testnet demonstration. It does not establish general-purpose AI certification or production deployment.