Towards Multilingual AI-Assisted Software Engineering

Abstract

India produces about 1.5 million Computer Science undergraduates each year, yet software engineering tooling remains predominantly in English. As the industry transitions to AI-assisted development, this English-centric assumption fails a significant portion of the global developer workforce. While AI and Large Language Models (LLMs) provide a multilingual foundation, their unconstrained application often suffers from semantic drift in low-resource Indian languages and may not consistently reinforce the English terminology commonly used in professional software development contexts. To address this gap, we propose Vocabulary Bridge: a terminology contract for bilingual code explanation. Unlike generic translation tools, this protocol aims at explaining code logic in the developer’s native language while explicitly retaining standard technical English terms. As early feasibility evidence, we present BhashaSE, a Visual Studio Code extension implementing this protocol for Telugu, Hindi, Tamil, and Kannada. By formalizing this approach, we aim to guide LLM generation to support the transition from native understanding to professional English proficiency, establishing a new direction for multilingual AI-assisted software engineering. All tool-related artifacts are open-sourced at https://github.com/rishalab/BhashaSE