Team10-KoEngage: Threads of tongue; Supervised Adaptation of LLMs for Korean-English Translation

Indian Institute of Science Summer 2025 DA225o Submission18 Authors

Keywords: Bilingual Translator, Korean, English, Deep Learning, Prompt Engineering, Supervised, SFT, Text Translator

Abstract:

We propose KoEngage, a bilingual translation system designed to convert text between Korean and English language seamlessly using modern transformer-based language models (LLMs).
This project leverages open-weight pre-trained models as the base model to build a robust translation engine.

To enhance model performance, we explore prompt-engineering techniques along with supervised fine-tuning, allowing the system to better adapt to translation
nuances and context-specific expressions. The combination of prompt-engineering and instruction tuning for weight optimization enables KoEngage to achieve better results quantified through known metrics,
contributing towards effective bilingual communication using deep learning.

Submission Number: 18