The goal of this project is to improve the accessibility and comprehension of scientific research papers for people working in the biomedical sector. Research publications are frequently written in technical language that is challenging for non-experts to understand, making it more difficult for them to stay current on the most recent research and use it in their work. This project intends to bridge the gap between technical research and practical application by simplifying complex articles and breaking them down into simple concepts. This will enable people in the biomedical industry to stay informed and use the most recent research to better their work.

What do we do?

The project provides a selection of tools made to increase the usability and accessibility of research articles. The initiative makes it possible for users to immediately understand the main ideas of a given document by distilling lengthy research papers into concise summaries and presenting them in slide style. Additionally, the project makes use of keyword analysis to give readers a deeper grasp of the subjects and ideas that were covered in the article. Users can also engage in a chat feature that provides answers to questions related to the research papers. Finally, the project offers paper recommendations based on the user's interests and reading history. All of these features work together to make research papers more digestible and increase their practical applications in the real world.

How we built it?

We have utilised pre-trained State-Of-The-Art Transformer Architecture models - BioLinkBERT (Base) and T5 for Question-Answering and Text Summarisation tasks. These models were determined after a very rigorous and thorough analysis as shown on the next slide.

Accomplishments that we're proud of...

Our end-to-end deployed system gives the user to interact and understand the document in a way never done before!

  1. KEYWORD ANALYSIS : Understanding the distribution of words across the document.
  2. TEXT SUMMARIZATION : Generating concise summaries of sections in the document
  3. SLIDE DECK GENERATION : Reproducing the contents of the document in the form of a simple, short and aesthetic PPT
  4. RECOMMENDATION SYSTEM : Providing suggestions to users based on relevance and upcoming latest trends.

Uniqueness of solution

Our solution is unique, different and novel compared to other similar solutions since it combines all the different tools under one roof. It addresses the biggest barrier to reading, understanding and explaining complex concepts/ideas to a wider public. The project has been built with industrial-level coding practices as well as effective collaboration (with tools like GitHub for Version Control, Drive for Document Collaboration). The Unique Selling Proposition of our system is its ability to harness cutting-edge Artificial Intelligence models with sufficient scalability at the moment (with the option to expand). However, this is not the end, merely just a beginning. There are tens of thousands of different methods (the number is growing!) to be applied and tested with every single passing moment..

What we learned

  1. Deployment using Streamlit package.
  2. Google Slides API documentation.
  3. HuggingFace transformer inference as well as pre-trained checkpoints.
  4. Parsing PDFs and understanding the implementation of BeautifulSoup and other libraries.

What's next for ArchMed

  1. Citation Networks for recommender system.
  2. Using Generative AI to produce explanatory videos.
  3. Containerisation (Docker/Kubernetes) to improve scalability.

About us

We are a diverse and equally qualified group of penultimate-year students from Nanyang Technological University, Singapore.

Built With

  • artificial-intelligence
  • deep-learning
  • deployment
  • google-cloud
  • huggingface
  • python
  • software-engineering
  • web-app
Share this project: