Cryptography and AI for adversarial environments.
Our research assumes the hard case: participants economically incentivized to cheat, permissionless infrastructure, and no central authority to appeal to. Systems that hold up there hold up anywhere.
Privacy-preserving computation
Our earliest federally funded work, under an NSF SBIR award, focused on hardware-accelerated fully homomorphic encryption for enterprise AI. With Flashbots we developed a Private Set Intersection protocol built on FHE, with applications from contact tracing to privacy-preserving auctions.
Verifiable computation
We work on zk-SNARK methods for verifiable queries and provable micropayment receipts, so that a result carries its own proof of correctness.
Economic mechanisms for agents
AutoAgora applied reinforcement learning to live query pricing on The Graph, negotiating costs that span four orders of magnitude. Pricing, bidding, and settlement between autonomous counterparties remain core research threads.
Agent identity and trust
Our current work gives agents decentralized identifiers, verifiable credentials, and reputation grounded in measured outcomes. This is the research feeding directly into Agentium OS.