print('{} is all you need!'.format('SumMed'))

Inspiration

It is frustrating to be a scientist in this world. At the rate of 2 papers poured into PubMed database per minute, it is frustrating to keep up with the latest in your field. Worse, besides reading through all these papers, you also need to make a nice presentation (weekly!) for your meeting (weekly!!!).

We believe that time should be spent more wisely. Researchers should spend their time on the project most relevant to them, not skimming through papers to pick up for their presentation. Therefore, we introduce SumMed - an AI-powered tool to shorten the time you need to create presentation, freeing you time to focus on what matters.

What it does

SumMed supports three main features:

  1. Drag n' Drop your PDF in, key information (author, summary, key words, tables and figures) out.
  2. Drag n' Drop tables and figures right into your PowerPoint slides.
  3. Interactive Chat on the content of the paper for smooth slide creation.

Challenges we ran into

We first want to create an end-to-end tool that generate PowerPoint slides straight away from PDF file. However, due to time constraint and technology underdevelopment, we settle for the next best thing: extract information from the PDF and present it in a PowerPoint-ready format for researchers.

Accomplishments that we're proud of

  • Implement a complete prototype of the idea with an intuitive UI.

What we learned

  • Prompt engineering and the capability of the LLM.
  • Web development.

What's next for SumMed (is all you need)

  1. Fine-tune Detectron model on hand-annotated researcher papers (~200) for smoother information extract.
  2. Integrate SlideMaker into theUI.
  3. Create an end-to-end pipeline going from PDF to PowerPoint slides.

Landhuis, E. Scientific literature: Information overload. Nature 535, 457–458 (2016). https://doi.org/10.1038/nj7612-457a

Built With

  • detectron2
  • flask
  • gpt-index
  • openai
Share this project:

Updates