Nexum | Devpost

Inspiration - Generative AI is Broken

With the rise of 10B+ parameter AI models like GPT-4 and Stable Diffusion, the need for high-quality data to train AI models has risen.

This leads to closed-source models using billions of data sourced from artists, creators, even people like yourself. Those who created the data have no way to benefit from the billions of dollars these models will generate. New entrants to the AI space are looking for more differentiating training data that is appropriately licensed for generate AI models.

Nexum connects these two underserved groups with a DeFi marketplace for training datasets.

Sidenote: I (David) work at You.com as a software engineer, so I have an understanding of how data-intensive large LLMs can be.

What it does

Nexum is a decentralized dataset marketplace built on the Ethereum blockchain and Jackal Protocol. It provides a secure and efficient platform for buying and selling datasets - enabling data providers and consumers to interact directly without intermediaries.

The marketplace leverages the decentralized storage solution provided by Jackal Labs to have an on-chain solution for reliable distributed storage.

Nexum empowers creators, artists, and even normal people to take ownership of their data and share in the profits of AI models. We also empower organizations to find diverse sources of data while ensuring that the data they train on is copyright compliant.

How we built it

We built the smart contract using Solidity and Hardhat, which provided some developer tooling/infra. However, we still found that to be lacking - especially since we were both new to blockchain technologies.

Users interface with the smart contract on a NextJS frontend. There, they are able to connect their wallets using MetaMask, and buy/sell datasets on the Ethereum Blockchain. All datasets that are created are zipped and uploaded to Jackal Protocol's storage service, with some internal APIs powering image previews (for latency).

Challenges we ran into

Maybe it was because we were new to developing and using smart contracts, but we found that there weren't many great debugging tools for Solidity contracts. A lot of error messages were quite vague and left us guessing/checking to find solutions.

One specific error when setting up the purchasing of datasets on chain left us stumped for hours. Eventually, after going through many chats with YouChat and digging through documentation, we finally figured out that there was an issue with how we were signing transactions using MetaMask. This was quite a blocker and stopped us from making progress on our non-MVP features.

Accomplishments that we're proud of

More than anything we are both extremely proud to have contributed and built something on the blockchain. Given that we knew exactly nothing about deploying a dApp - let alone writing one - 15 hours ago, this project was a great personal accomplishment.

What we learned

From a technical aspect, our biggest learning came from understanding how the blockchain works from a development perspective. Being able to understand the code behind a smart contract and write the code in Solidity made it easier to understand how larger industry projects work and the technicals associated with them. Additionally, using Jackal introduced us to decentralized storage and how it can put the power back in the hands of users. With Jackal being such a new platform being able to work through technical challenges alongside the Jackal team taught us how to work with what we have and instilled a scrappiness within our team.

What's next for Nexum

Better design for multiple file types. An on-chain protocol, and slightly better loading states/UX

Jackal Protocol Pros/Cons

Extremely fast even though the protocol is still in early stages
Relatively easy to set up, equivalent to something like Firebase storage or S3
Documentation was sparse (lack of examples, etc)
Protocol could have better states exposed for each transaction (finding host, uploading, hashing, etc...) instead of just one loading state