ETHGlobal NYC was a useful compression test: could a small team turn decentralized AI inference, GPU providers, wallet payments, and a terminal-first developer interface into something demoable before the window closed?
The raw project vault had too much operational noise for the public site. The public lesson is sharper: hackathons reward teams that separate the demo surface from the task surface.
What We Built
The project explored a decentralized inference network where model providers could expose compute and consumers could pay per request using crypto-native payment rails. The most public-facing artifact was a CLI/TUI demo: a small surface that made the network feel concrete without requiring every backend assumption to be fully mature.
Constraint
The hardest constraint was not code volume. It was decision bandwidth. A hackathon creates constant pressure to research partner tracks, make technical choices, produce a pitch, build a demo, coordinate teammates, and keep the story coherent.
If all of that becomes public documentation, the project looks chaotic. If none of it is captured, the learning disappears.
Reusable Idea
Keep three layers separate:
- Demo layer: the smallest experience that proves the idea.
- Decision layer: the tradeoffs, architecture choices, and partner constraints worth remembering.
- Task layer: the messy execution board that should usually stay private.
The public second brain should publish the first two layers only after editing. The third layer can be useful internally, but it rarely teaches an outsider much.
What Carries Forward
- Terminal interfaces can be excellent demo surfaces when the audience is technical.
- Payment primitives need to be explained through user flow, not only architecture.
- Public postmortems are more valuable than raw hackathon task lists.
- The next public artifact should compare crypto payment rails for AI inference and explain which part of the stack needs decentralization.