Researcher App

An app that aims to organize your research: A Researcher with a Paper Based Approach

Author: @stepp1

🚨 DISCLAIMER:

This is a WIP. I worked on it for two weeks and wanted to have an MVP fast. When I started it, I didn't have any experience with Streamlit nor LangChain.

Any feedback is welcome!

Currently, the app supports the following features:

• 📚 LibraryChat: a chatbot that helps you understand a collection of papers. Powered by 🦜️🔗 LangChain, 🤗 Hugging Face Instructor, and 🤖 OpenAI.
• 🔎 PaperExplorer: a tool to explore papers in your library. Powered by 🤗 Hugging Face Instructor and Plotly.

Table of Contents

• Status
• Installation

Status

• TODOs:
  • Add: PaperChat: a chatbot that helps you understand a single paper
  • Improve: Full Text Extraction
  • Improve: Data Pipelines
  • App: Decide if re-implement in a more flexible framework (e.g. Flask or FastAPI)
  • App: Include pdf viewer when clicking on a paper
  • Testing: add tests for data processing, embeddings, etc.

Installation

You will need a .env file with the required keys: HUGGINGFACEHUB_API_TOKEN, OPENAI_API_KEY.

1. Clone the repository

git clone git@github.com:stepp1/research-app.git

2. Install the dependencies using conda/mamba

conda env create -f environment.yml

3. Activate the environment

conda activate researcher

4. Run the apps

chmod +x run.sh
./run.sh

Remember to forward the port for streamlit if you are running it on a server!

Dataset

We provide a dataset.json file stored at researcher/data/ that contains metadata for the default papers.

Full dataset is currently hosted on Zenodo: https://zenodo.org/record/7653458

More information about the dataset can be found here.

Download the dataset

• Download the dataset.json file only:

curl -L https://zenodo.org/record/7653458/files/dataset.json -o researcher/data/dataset.json

• Download the dataset.json file and the images:

curl -L https://zenodo.org/record/7653458/files/data.tar.xz | tar -xJ -C researcher/data/