Proving the work,
on machines you do not own.

Gensyn's premise is that machine learning compute should not be controlled by a handful of providers. Its protocol lets independent operators contribute GPU power for model training, coordinated in a way that is permissionless and neutral.

The company also does genuine research alongside the engineering — verification systems, and work on training large models without all-reduce. This is not a product with a novel UI; the novelty is in the protocol itself.

If anyone can contribute compute, anyone can lie about it. A node can claim to have trained a model and return plausible-looking garbage, and there is no central authority to appeal to. So the protocol needs to make correctness provable rather than trusted.

That has to work while nodes are heterogeneous — consumer GPUs alongside data-centre hardware — and while participants join and leave at will. Identity, verification and reward all have to be handled without anyone in charge.

• R1Secure, verifiable infrastructure for managing node participation
• R2Trustless validation of results returned by untrusted nodes
• R3On-chain identity, verification and reward mechanics
• R4Orchestration across heterogeneous GPU hardware
• R5Off-chain execution coordinated with on-chain settlement
• R6Node metadata and contribution logging that stands up to dispute

We built across both layers — smart contracts handling identity, verification and rewards on a custom roll-up, and Node.js microservices orchestrating the off-chain execution that the contracts settle against.

Why this work is unusual: almost all of the difficulty is correctness, not interface. There is no user to forgive a rough edge — a verification bug means the protocol pays for work that was never done.

• 01Blockchain layer. Custom Ethereum roll-up with smart contracts for identity, verification and rewards.
• 02Backend services. Node.js microservices for orchestration and off-chain execution.
• 03Data layer. PostgreSQL for node metadata and contribution logs.
• 04Compute layer. Coordination across heterogeneous GPU nodes, from consumer cards to data-centre hardware.
• 05Verification. Systems supporting reliable, verifiable evaluation of machine learning work.