Verbally – An AI-Powered Companion for Mastering English

Abstract

In today’s digital age, English learning remains a fragmented process, requiring students to rely on a variety of online courses and applications to develop different language skills. However, these solutions often lack clear feedback and structured learning paths, limiting their effectiveness in providing an adaptive learning experience. This project proposes Verbally, an AI-based English learning system designed to transform traditional e-learning platforms from simple content delivery tools into intelligent systems capable of assessment and personalized feedback. The key objective of the project is to create an easy to use, complete, scalable, and self-paced. It should be able to test a learner's skills in all of the core English skills, such as vocabulary, grammar, listening, speaking, writing. The framework is constructed based on a robust three-tier client-server model. React is used to create the front end, which is deployed into Vercel, the backend is constructed using NestJS and run on Railway. Aiven is used to manage PostgreSQL database. In accent based speech assessment module, it uses Deepgram Speech to Text and the Montreal Forced Aligner to provide precise, phoneme level and accent aware feedback. It utilizes the Google Gemini API, which is a generative AI model, to automatically create lesson content, writing prompts and test questions. The common European Framework of Reference of Languages is utilized in the platform to put students to A1-C2 proficiency levels. A Level-Up module allows learners to advance to the next level by taking combined vocabulary, grammar, writing and speech tests. A hybrid python microservice architecture is used for speech scoring pipeline, which uses a lot of processing power. The system also has a real-time messaging feature, and it allows learners to be socially connected. The system aims with response times under a second, 99 percent up-time and capability to service at least 100 number of concurrent users dependent on its beta. This demonstrates that it is a suitable and practical alternative to discontinuous English learning solutions