Inspiration
There have been many times that more than a few of our members have had to read long research papers filled with technical jargon. The complexity of these papers can cause confusion for researchers, and we want to help change this trend.
What it does
RTCL ingests a PDF file of a research paper, and outputs a powerpoint (pptx) slide deck that summarizes that paper using the co:here NLP AI models.
How we built it
The models were trained using presentations from Dr. Nesterenko's seminar group. These presentations were scrapped and used to train the model for its own text generation, and these summaries were labeled for the classification model.
Challenges we ran into
Our group had some issues working with web apps and ingesting the data. We also had problems getting the powerpoint off of the server and into the users machine in an easy way. We are sure that there are easier solutions than what we have implemented, but we just had to spend our time elsewhere.
Accomplishments that we're proud of
Accomplishments we are proud of is our pipeline that we have created using the co:here API and pdf files. We are confident that in a setting where we are given more time, better models could be trained and tuned, and our output could be even better.
What we learned
We learned a lot about the co:here API, as well as web development using python. NLP was new to all of our group members, and everyone learned something about the models during the hackathon.
What's next for RTCL
RTCL might not see some improvements for a little while, but the software is definitely positioned for easy improvements. If someone with more web Dev knowledge was interested, or if the group revisits once we gain some more knowledge, we are sure that the product could be even better.
For the server hosting our index, go here
Log in or sign up for Devpost to join the conversation.