The Client & the Challenge
An international education company, one of the largest operators of kindergarten-to-grade-12 schools in the world with a network of over 80 schools in over a dozen countries was looking for a Q&A bot for educational coursework queries. The customer also wanted to get quizzes, MCQs, and descriptive questions generated automatically as per student’s’ grade syllabus.
The education industry comprises establishments whose primary objective is to provide education. Educational institutions, in a significant way, contribute to the development of educational resources. They are entrusted with providing the talent pool for other industries, which is a critical factor when competing in a global economy. Technology is being integrated into every aspect of the education industry. Educators can now use other modes of delivery to teach and communicate with students in the classroom. AI can not only customize education and training courses for students but also assist teachers in their key tasks and responsibilities.
In today’s highly competitive and advanced education sector it becomes imperative to be available 24*7 for rendering assistance to students. The customer wanted to establish a Q&A bot for students’ educational coursework queries. Also, year after year, coming up with new sets of questions from the coursework was an uphill task for teachers and faculty. So, a solution generating and delivering unique and different types of questions automatically would help educators in this aspect. They primarily wanted to solve two primary problems: 1. Questions get answered based on students’ grade syllabus and the required answer to be generated along with the relevant image. 2. Also, educators should be able to generate new and distinct fill-in-the-blanks, descriptive, and multiple choice questions for students automatically.
AI can not only customize education and training courses for students but also assist teachers in their key tasks and responsibilities.
We used a blend of vision AI and Deep Learning to solve the customer's challenge. Here is a breakdown of the steps we used:
Step 1: Selection and Usage of large language models
To solve the problem, the primary task is to make the software system understand the language; hence, we selected the GPT-3 large language model. Large language models are computer programs that open new possibilities for text understanding and generation in software systems. Large Language Models (LLM) were selected to achieve the goals that use machine learning algorithms and can recognize, predict, and generate human languages on the basis of extensive text-based data sets and can solve NLP problems on zero-shot mode and query the model with a prompt. GPT3, Generative Pre-trained Transformer 3, is an autoregressive language model that uses deep learning to produce human-like text.
Step 2: Syllabus provided as context to the model
The syllabus was provided as the context to the large language model. At the same time, to provide content moderation for kids' safety, separate models were trained based on kids’ safety filters. Also, custom models were trained and built for question generation for some specific subjects and a certain category of questions.
Step 3: Transfer the model output to the school dashboard
AI system was integrated with the client’s platform providing different dashboards for students to solve their queries and teachers to generate automated questions. Educators were able to create Fill-in-the-blanks, MCQs, and descriptive-type questions automatically. Also, students could quickly get their questions answered automatically from their dashboards.
- Increased Student Engagement with multiple subjects support
- Provides Students Instant Help with syllabus based responses
- Acts a Teaching Assistant
- Solution deployed for 130000 students
- Response within 1 sec