Built with:

Edulytica

Description

The purpose of the application is to automate the analysis of scientific and educational documents in the context of research works using LLM (Large language model, large language models) to reduce the time and intellectual costs of teachers. The result is two LLMs, trained on specially collected data, capable of summarizing the text of a large document and revealing whether the stated goals and objectives of the work have been achieved.

Features

• Algorithms for summarizing large texts and evaluating the achievement of stated goals and objectives;
• Fast APP is an application for interacting with trained models;
• Separate models have been trained to summarize and assess the achievability of goals and objectives;
• Datasets have been prepared for training models, separately for summarization, separately for goals and objectives.

  Please help us improve this project, share your feedback with opening issue!

Installation

1. Clone the repository

git clone https://github.com/LISA-ITMO/Edulytica.git

2. Activate venv

source ~/PyProject/Edulytica/api_venv/bin/activate

3. Install requirements

pip install -r requirements.txt

4. Start Application

python3 src/edulytica_api/app.py

5. Activate Celery

celery -A src.edulytica_api.celery.tasks worker --loglevel=info -E -P gevent

6. Run npm

npm start

7. Run Celery task

celery -A src.edulytica_api.celery.tasks flower

Getting started

First, you can familiarize yourself with the examples in JSON format of the system's responses to the test sample of works.

When you have managed to launch the service, you can send the documents yourself and get acquainted with the results of their verification!

Documentation

Details of the documentation can be found at the links below:

• algorithms - part of the task of analyzing the text how much it is necessary to change the source text (which is written by AI) so that AI recognition systems do not recognize AI in this text;
• data_handling - an auxiliary module that stores parsers of data and documents for generating datasets;
• edulytica_api - this module stores the source code of the web service;
• extracting_rules - This module is devoted to an experiment with extracting design rules using LLM;
• rag - Package for an experiment with semantic search, kNN and the mBERT model are used.

Code documentation is available at the link.

Requirements

For more information, see the file requiremets.txt.

Contacts

Our contacts:

• Martsinkevich Viacheslav, slavamarcin@yandex.ru;
• Tereshchenko Vladislav, vlad-tershch@yandex.ru;
• Aminov Natig, natig.aminov@gmail.com.

Publications about Edulytica

We also published several posts devoted to different aspects of the project:

In Russian:

• Edulytica: LLM-ассистент для проверки научных работ - Scientific Open Source Meetup №8, October 2024.

Authors

Tereshchenko Vladislav\ Martsinkevich Viacheslav\ Aminov Natig\ Mischenko Maxim\ Bogdanov Maxim\ Dvornikov Artem\ Laptev Egor\ Sinyukov Lev\ Marakulin Andrew