Lilypad Network

Compute has become one of the quiet gatekeepers of science. If you need serious GPU time and don't happen to work somewhere that owns a cluster, you wait, sometimes as long as eighteen months, while your questions sit idle. Lilypad Network attacks that bottleneck head-on. It's a decentralized compute marketplace, an Uber for GPUs, connecting people who own idle graphics cards with people who need them, in a trustless and verifiable way.

Verifiable is the hard word in that sentence. If a stranger's GPU runs your job, how do you know it actually ran it, and ran it correctly, instead of handing you back a plausible-looking guess? As Head of Research, I co-authored "Validation of GPU Computation in Decentralized, Trustless Networks" (arXiv 2501.05374) on exactly this problem. Re-running the computation to check it fails, because GPUs aren't perfectly deterministic. Specialized secure hardware isn't always available, and the heavy-duty cryptography that could prove it is too costly to run at scale. Our approach combines model fingerprinting, semantic similarity, and GPU profiling into a binary reference model with ternary consensus, so the network can catch a dishonest result without redoing the work.

Then I wanted to see the compute do something that mattered. I launched the Altruistic AI Agents series, building AI agents aimed at real-world good. The first episode wired up a four-node agentic oncology pipeline with a Cedars-Sinai researcher: one node searching the literature, one reading it, one reasoning about the oncology, one designing candidate proteins, all running on Lilypad's decentralized compute. Someone very close to me went through cancer and didn't survive. This work is personal.

Lilypad is where Lattice Protocol's trust layer meets a real clinic. The theory of verifiable compute only earns its keep when a patient is on the other end of the result, and making that compute trustworthy is the part I intend to keep getting right.