CuriosityAI

Our Landing Page.

Inspiration

Lot's of students and professors in academia perennially struggle with finding new and novel ideas to research. We want to bridge that gap by finding novel research ideas, and also help them get started on it by quickly bootstrapping them.

What it does

We take user input, which is keywords of ordinary or well-researched topics in academia, and industry, and get research papers/articles, etc. from various sources and find relatively less researched or zero-research area topics.

How we built it

We get the research papers/ articles related to the keywords in the user input. We perform vector embeddings on the abstracts, and the titles of the papers/articles. We then reduce the dimensionality to be able to plot a 3D scatter graph to better represent our initial space. We want to take the negative/sparse areas of the spatial representation, as they represent a relatively less explored space, and hence, might correspond to niche and novel research ideas. We also use AI agents to further get our userbase started on implementing/understanding/further researching on the idea. We have an AI agent that can help set up a basic working running coding environment, and another that provides relevant references to further research the topic, and another agent that determines if the idea is feasible or not, and if it is an ethical idea to pursue or not.

Challenges we ran into

We ran into quite a few challenges while building this. One of them is finding the right embedding model, according to our use case, and also our current hardware infrastructure. We have to take care of dimensionality, and make sure it doesn't become a crutch in finding novel ideas. We also run into problems while creating and deploying AI agents using Fetch.ai. Deploying and integrating all the agents with our core idea proved to be a challenge that we bested.

Accomplishments that we're proud of

We made a beautiful 3D graph that shows our existing spatial representation in graphical format. We were able to generate a tangible novel research idea from the negative vector space. We've successfully implemented and integrated multiple AI agents into our workflow.

What we learned

We've learnt that spending time on integration is just as important as spending time on the implementation. Also, we've learnt that it is worth strengthening the fundamentals as that would help in making tough decisions during architechture, and implementation plannings.

What's next for CuriosityAI

We are going to push the envelope on how AI agents are going to be used in conjuction with other everyday functions, and tasks. We look forward to exploring other more accurate, and even novel techniques to find out interesting and novel research ideas from exisiting ideas or a set of existing ideas.