Encrypted fitness-data pipeline on TACo threshold access control with condition-based decryption, giving users cryptographic control over how their own training data is delegated to third parties. Built under embedded power and network constraints on ESP32 hardware with over-the-air firmware updates.

Built for Hobbs, seeded by a $30,000 first-place Zahn Innovation Center grant in 2022. hobbsapp.netlify.app

You cannot tell which model is actually running behind an AI agent, and a provider has a commercial reason to swap in something smaller or quantized.

SPOT is a session-level protocol. The provider commits to their exact weights once and publishes the hash. Every response afterwards carries a signed receipt bound to that commitment, and a sampled subset carries a zero-knowledge proof that those weights produced that exact output, without revealing them.

Demonstrator · Source

A proof pipeline over multi-hour GPU workloads on NASA-ISRO NISAR radar data: a RISC Zero zkVM guest re-executing a Fiat-Shamir residual check, wrapped in Groth16, with Intel TDX attestation over the off-chain engine.

Then I audited it, and the audit is the useful part. The Fiat-Shamir seed committed only to public inputs, so the sampled column set is computable before choosing what to publish. I built a working forgery: the certificate verifies, every check passes, and 999,744 of 1,000,000 output columns are wrong.

A self-hostable, IPFS-backed stack giving Zora-based NFTs guaranteed data availability, with a hardware wallet component. IPFS nodes ran garbage collection that made NFT media intermittently unavailable, so the token resolved and the artwork did not.

Built at ETHGlobal NFTHACK 2021. Source · Showcase · Demo

I teach a weekly developer-club series as a Bitshala Educator Fellow, working from one-way functions and proof-of-work through secp256k1, Schnorr signatures (BIP-340), and threshold signing with MuSig2 and FROST.

bitshala.org