Web3 Research

• 

  Web3 x A.I.

• 

• 

• 

Inspiration

I was inspired by a problem I had: Which influencer should you trust? And how can I get better information from these people faster and more accurately with the help of A.I.?

What it does

The application takes any user handle on Twitter with a date range and specific token, processes the data and returns how accurate their sentiment about that specific token has been. The application also takes the top 'n' performing assets over 't' time and finds users who tweeted about the asset before it became popular.

The other side of the application is a non-custodial portfolio management tool that allows a user to create a follower account that they own and then has that follower account take any set of instructions and execute those instructions.

The long-term goal here is much larger than these features. I have access to a substantial amount of vram to train our own AI models to contribute to the above in meaningful ways. Not only can AI consolidate the information into more digestable signals, but it can also send instructions to the follower account. The long term goal here is to have an entirely personalized A.I. assistant that acts as a layer of abstraction on top of web3 technology.

How we built it

There are several components to this application- We have the sentiment analysis which is using a modified BERT model(we have our own proprietary one we have been training as well) We have python scripts processing the data and sending the alpha insights and sentiment analysis to a front end implementation the front end is probably the least interesting as we wanted to focus on getting the hard part done first(there are a lot more front-end developers than smart contract or AI developers) but it does demonstrate that data from the python flasks is being sent to the front end with some graphical outputs.

Challenges we ran into

The biggest challenge we ran into was time management. we anticipated the A.I. side of the project to be more time-consuming, but more time was spent on the smart contract implementation. The smart contract testing is what easily took the most time. We only got 4.5 of the 7 tested and were not able to connect them to the sentiment price or alpha insights yet for trading signals.

Accomplishments that we're proud of

This was a monumental project with multiple moving parts that all needed to work together. To get even half of the way to where it is now is amazing.

What we learned

We learned a lot about testing and found we are quite competent in the A.I. space now. Practice makes perfect!

What's next for Web3 Research

We are just getting started with this project. We are looking for a chain to call home, but need some funding first. So we will keep building and looking for a sponsor to help us make this a reality faster.

Built With

• ganache
• python
• react
• solidtiy
• truffle
• typescript