Extract - Major Project

Table of Contents

1. Problem Statement
2. Solution
3. Directory Flow
4. Working
5. Deep Learning Modes Used
6. Tech Stack Used
7. Future Initiatives
8. External Lins

Problem Statement

There are many sites available over the internet to find answers for your problems/bugs like Chegg, StackOverflow, but none of the sites offer peer to peer answering in form of images or digital media. The biggest perk of this sharing is no requirement of typing and answering question. It is most helpful in the cases of academic questions. As most of the questions students of any corner of the world knows in the form of his/her notes, but don't bother to type and answer the question.

Solution

Extract is solution for this problem, which offers three important solutions to world at one place:-

1. Storage to store the notes in the form of images which solves the problems of hard copy notes.
2. Extracting necessary data and questions from those images to provide answers globally available.
3. Questioning as in StackOverflow where users community can manually link questions asked to the answers/images.

Directory Flow

Extract
|--- Backend
|    |--- middleware 
|    |     |--- Contains middlewares for backend
|    |--- models
|    |     |--- Contains all schemas
|    |--- routes
|    |     |--- All routes
|    |--- index.js
|    |--- db.js

|--- extract
|    |--- public
|    |    |--- Public files
|    |--- pages
|    |    |--- All pages of website
|    |--- components
|    |    |--- All components of website (ReactJs part)

Working

• Working of Extract involves Login/Signup as a user.
• Now you are eligible to create as many folders as required and add your notes to the folder.
• Also you will be given an option to choose whether to make the folder public/private.
• Public folder means the data will be extracted from the images uploaded in that folder to make them available gloabally.
• Now the public folder images data is extracted filtered using AI/ML models and then question is extracted out of the image.
• The question with its image answer is now available globally to the users.
• One another feature of Extract is asking questions, if question is not answered yet or can't match the relevant text, then users community can link the question to the image.

Deep Learning Modes Used

• Firstly the model used for OCR of image is docTR. It is pretrained and developed by Open-source community. It required all the requirements mentioned in requirements.txt.
• Install all requirements then you are ready to go.

Why docTR is used?

• I used docTR model because of very high accuracy as compared to any other OCR and it can very easily judges number and symbols which are essential in high studies.

What is the working of Model?

• Firstly any user uploads their notes in their respective folders which have an option of keeping it private/public.
• If folder is public the image uploaded in first OCR using docTR and then complete text is send to another Deep Learning model.
• The deep learning model was created by me from scratch using LSTM, NLP, Convolutional layers, Embeddings.
• Now the model, filters out the questions from image and store it!

Tech Stack Used

--> Node JS Libraries

npm install bcryptjs

npm install body-parser

npm install cors

npm install dotenv

npm install express

npm install express-validator

npm install jsonwebtoken

npm install mongoose

--> Next.JS/React.JS Libraries

npm install react-dom

npm install mdb-ui-kit

npm install -d autoprefixer

npm install -d postcss

npm install -d tailwindcss

Future Scope

External Links

• Website Link
• Backend Api
• Deep Learning Model