Fill In The ___

• 

  An example of our python ML fill in the blank model working.

Inspiration:

We were inspired by the games among us, scribbl.io and mafia.

What it does:

Our project uses ML to generate text to fill in the blank between phrases. One player plays as the imposter (uses the generated options provided) and others have to single him out. The imposter is incentivized to fit in and convince the others that his sentence is not actually written by the AI. The players were supposed to be able to connect in rooms and have voice + text chats but certain technical aspects with socket.io prevent us from getting that implemented.

How we built it:

We created a react app for the frontend, and react-bootstrap components for the UI Our backend was an express server.

Challenges:

We failed to correctly implement the socket.io dependency, resulting in an app that doesn’t support multiple users. Additionally, integrating our backend with the python AI for NLP was a challenge.

Accomplishments:

First, we are proud that we haven't given up and are working on this project until the last minute. We are proud that we have not decided to quit despite many challenges. We are also proud of our AI. We didn't expect to get a very accurate and coherent model for multi-token mask filling.

What I learned:

We learned a lot about recurrent neural networks and transformers. We gained a much deeper understanding of the different tasks and areas of study in NLP.

What's next for Fill In The ___:

We will try to get our chat/call room to work properly in our web app. We will also explore the option of making a standalone chat-room Python.

Built With

• javascript
• node.js
• python
• pytorch
• socket.io