DAIM (Decentralized AI Marketplace)

What's the biggest problem with AI? Transparency and Complexity. This app fixed the DAIM (dayum) problem by making a decentralized AI artifacts sharing platform

Comment 1

Inspiration

“Oh, did you look at Open AI’s new model?” “AI is the future; agentic workflows are the future”. We hear this almost every other week in news nowadays, but let’s take a moment to analyze the “glorious future” we are heading towards. There are various problems that plague AI currently; some philosophical like whether future more advanced versions of AI would be considered sentient or is it sentient now, some legal like Open AI being guilty of copyright infringement while training its data. The two biggest problems that plague modern artificial intelligence: difficulty in understanding the model and its decision and the large model’s functioning being gatekept by large corporations. DAIM solves that problem by creating a decentralized platform to share AI models and datasets.

What it does

DAIM is an acronym for decentralized AI marketplace with provisions for non-technical users to effectively participate in keeping AI open-source and for the community. Some of the features of DAIM are:

  • Wallet based authentication and transaction signing to verify uploading user's identity
  • AI model listings highlighting important key performance indicators like accuracy and parameter count
  • Dataset listings with key performance indicator like rows, columns, and type of dataset available to download.
  • AI Chatbox interface with contextual awareness to help user's solve their problems effectively.

How I built it

The platform was built using modern typescript and javascript frameworks in conjunction with blockchain technologies. The model code is stored with pinata IPFS and the datasets are stored on Autonomys on-chain Auto Drive network. The reason for this tech stack is that model source code is not the most sensitive information, since there are multiple ways to represent an idea through code. However, machine learning data is often hard to procure and easy to mishandle. As a result, the model source code is stored on IPFS while the dataset is stored on the Autonomys Auto Drive.

Challenges we ran into

The biggest challenge I ran into was integrating the Autonomys Auto drive SDK. The documentation was vague in some crucial points which caused problems with implementing the Autonomys features. I was in touch with @cryptohaki from the Autonomys team, but ultimately I ran out of time before perfectly integrating Auto Drive.

Accomplishments that I am proud of

This is my first major blockchain project. Apart from simple code snippets and one-liner smart contracts, I have never used these many blockchain technologies together.

What I learnt

I learnt how to apply web2 software engineering principles to blockchain technologies and force myself to think how to decentralise the application. Understanding that it does not always have to be google, meta or other big companies to create reliable platforms to share knowledge was a great refreshing realisation, and that the community can work together to ensure knowledge remains in the community's hands.

What's next for DAIM

I plan to solidify the idea further and make it secure. There are still some concerns about security that I was not able to address in the limited hackathon timespan. The ideal goal for DAIM would be to be a large eco-system similiar to NPM, using the CLI to seamlessly download AI resources.

Built With

  • autonomys
  • gemini
  • ipfs
  • metamask
  • nextjs
Share this project:

Updates