shristighimire/Language-Translation-Model-for-Underserved-Communities-

A GPT-3-based translation system for Nepali, Mongolian, and Tigrinya, fine-tuned with RAG to improve accuracy and deployed on AWS SageMaker for scalable, real-time translations. # Language Translation Model for Underserved Communities ## Overview The **Language Translation Model for Underserved Communities** project aims to bridge communication gaps in underserved communities by developing an accurate translation system for **Nepali**, **Mongolian**, and **Tigrinya**. By fine-tuning **GPT-3**, this model enhances translation capabilities and makes it easier for individuals in these communities to access critical information in their native languages. The project leverages **Retrieval-Augmented Generation (RAG)**, where relevant translation prompts from these languages are used to improve the model's performance. The fine-tuned model was deployed on **AWS SageMaker** to improve scalability, reduce latency, and support high request volumes, ensuring that the system can handle **1,000 requests per minute**. ## Key Achievements - **Fine-tuned GPT-3**: Improved translation accuracy by **63%** for Nepali, Mongolian, and Tigrinya languages by leveraging custom datasets and RAG-based prompts. - **AWS SageMaker Deployment**: The model was successfully deployed on **AWS SageMaker**, which resulted in a **40% reduction in processing time** and ensured the system could handle up to **1,000 requests per minute** during trials. - **Translation Efficiency**: The fine-tuned model provides faster, more accurate translations that can be used by underserved communities to overcome language barriers. ## Features - **Fine-Tuning with Custom Data**: The model is fine-tuned using data from Nepali, Mongolian, and Tigrinya language pairs to improve translation accuracy. - **RAG Implementation**: The project uses **RAG files** to augment the translation process, making it contextually aware and more effective in translating less-common language pairs. - **AWS SageMaker Deployment**: Scalable deployment ensures quick processing of high-volume translation requests, reducing latency and improving user experience. - **High Throughput**: Designed to handle **1,000 re …