Abstract:
ZLearn is an online tool for IELTS preparation that offers preparation resources for all four IELTS test modules: Reading, Writing, Speaking and Listening. It's built on the front end using React, and the back end using Node.js with Express and MongoDB. For Writing and Speaking, we integrated Google Gemini 1.5 Flash and used it to analyze student work and provide feedback based on the official IELTS marking criteria for the four parts of the Writing and Speaking paper, which are Task Achievement, Coherence and Cohesion, Lexical Resource, and Grammatical Range and Accuracy. The Listening module presents audio waveforms and allows for a variety of question types such as gap fill, multiple-choice, map labelling, note completion and matching questions to be timed. All Listening and Reading answers are submitted and assessed on this, and results are displayed on a dashboard. There are different interfaces for students, teachers, and administrators of the platform. Students view their own score history, teachers add content and monitor students' progress and admins manage users. All screens are in dark mode and Login uses JSON Web Tokens. We compared the performance of the Writing scorer on essays with known band scores and calculated a mean absolute error of 0.3 band. This is within the limits of disagreement that are common between two human IELTS Examiners. It's hosted for free on the free tier with Render.com and MongoDB Atlas for storage. This enables a more efficient and scalable way to assess the proficiency of the language being spoken in the video. This allows the React website to evaluate the language level of the video's spoken content more efficiently and at scale, using NLP and Google Gemini.
