obrigado.cc

Inspiration

Near is aiming to be the Operating System for an open web with the on-chain frontend (BOS) and backend (WASM/zkWASM). The missing piece is a way to process heavy computes such as training a LLM or analyzing DNA data.

Problem

This missing piece in the near ecosystem forces users to rely on external chains or centralized services for heavy computational jobs, leading to potential inefficiencies and a disjointed user experience. The cost of training GPT-5 is measured in billions of USD. This high cost of iterations is slowing down and delaying the progress. Near has the potential to bring the cost of decentralized training down by breaking the monopoly on compute marketplaces set by few cloud providers.

Solution

Obrigado will allow seamless execution of heavy computational tasks, distributed across multiple computational resource providers (Task Runners), ensuring efficient processing and integrity of results, all within NEAR. Computes happen off-chain and book-keeping like job queue handling, result validation, node reputation, slashing and balances happen on-chain.

Target User / Market

• Entities ranging from independent researchers to large organizations seeking cost-effective, decentralized solutions for processing large batch computational tasks.
• Anyone with a device and internet connection who want to monetize idle computational resources.

How we built it

Our development process involved creating a user-friendly interface for task submitters to easily upload tasks and set bounties, and a seamless system for task runners to download the software to contribute their idle computing power.

Challenges we ran into

The biggest challenges lay in ensuring seamless interoperability within the NEAR ecosystem and maintaining a balance between on-chain efficiency and off-chain computation reliability. We also faced hurdles in designing a user-friendly interface that catered to both tech-savvy and novice users. Ensuring the security and integrity of the computations and creating a robust system for managing node reputation were other significant challenges that we will have to continue to work on.

Accomplishments that we're proud of

Obrigado.cc will be the pillar for heavy computing within the NEAR ecosystem. By integrating directly with NEAR's existing infrastructure, Obrigado.cc offers a seamless experience that enhances the overall value proposition of NEAR and its dApps.

What's next for obrigado.cc

• [ ] Will use docker to prevent damages from malicious jobs
• [ ] Handle non-deterministic jobs
• [ ] Implement something like Lit Protocol on Near to protect results being taken by public without work.
• [ ] Check the properties of the result instead of result equivalence (by hash)
• [ ] Federated learning
• [ ] Define the job to execute as URL to GitHub repo to clone and execute

Built With

• bos
• canva
• figma
• javascript
• jutsu
• near
• python
• react
• rust