Feature Proposal for EduAid

[abhyudayhari]

Structured Question Paper Generation with Marking Scheme and Model Answer Script

Current Functionality: EduAid currently supports generating various types of questions, such as short answer questions, multiple choice questions, and true/false questions, based on the input context provided as a paragraph or chapter content.

Proposed Feature

Enhance EduAid to support the following advanced functionalities:

1. Structured Question Paper Input: Allow users to upload the structure of a question paper in formats such as Google Docs, PDF, etc.The structure may define the types of questions (e.g., long answer, match the following, assertion reason etc) and their distribution in the paper.

2. Chapter Content Input: Enable users to upload the chapter content as a PDF or provide a link to the chapter. For multi-chapter documents, allow users to specify the mark distribution for each chapter.

3. Question Generation Based on Marking Scheme: Generate questions according to the specified structure and marking scheme. Ensure that the total marks of the generated questions match the given distribution.

4. Model Answer Script Generation: Along with the question paper, generate a model answer script. Provide point-wise mark distribution for each question in the model answer script.

5. Reference Linking: Include references for each question and answer, indicating the exact section in the chapter from which the question was derived.

Implementation

1. Input Handling: Read the uploaded Google Doc or PDF question paper structure using Python modules like PyPDF2. Extract key information such as question types, number of questions, and marks allocated. Implement OCR processing for scanned syllabus documents and convert them into structured text. Create a vector database from the extracted syllabus content for efficient context retrieval.

2. Question and Answer Script Generation: Fine tune different LORA adapters for different question patterns, and when those type questions are to be asked these adapters are loaded onto a base model LLama 1.3B/3B model and generate the output, this structure allows for modular and efficient question generation. This also also allows for a chatbot like interface where the user can change the questions or instruct the base model to change as per the user's demand. For content retrieval, a RAG pull to the vector database would be done for better reference to the user.

3. Paper Generation: All the generated questions and corresponding answer scripts would be dumped onto different PDFs or google form or doc format given by the user so the end user can use them directly without changes using python module like FPDF.

[dhruv-k1]

Hi @abhyudayhari , I'd like to work on this issue. Can you assign it to me?

[abhyudayhari]

Hi @dhruv-k1 I am not a maintainer of the project, Plz have a look at the discord channel for the updates

[abhyudayhari]

@Roaster05 @Aditya062003 I would love have your opinion on this