The scope of Ghanaian Language; Twi datasets is insufficiently diverse, hindering a multitude of applications. To address this, we are curating a comprehensive corpus from literature, media, and synthetic data, covering formal, informal, and conversational registers. Using advanced data collection methods and rigorous validation, including transformer-based models and expert human verification, we ensure a high-quality data repository for robust NLP applications.

An ensemble approach to enhance efficiency and interpretability in Biomedical AI through majority voting. This project focuses on equipping large language models (LLMs) with multimodal capabilities through feature alignment using a vision encoder. Additionally, we are experimenting with synergistically combining language (LLM) and vision (ViTs) models to to enhance interpretability in medical diagnosis.

This research project focuses on enhancing the reliability of Residue Number System (RNS) arithmetic operations through the development of effective fault tolerance techniques. We also proposed a novel integrated fault-tolerant technique that combines redundant residues, modular checksums, and Reed-Solomon codes for error detection & correction. This integrated approach aims to improve fault coverage and reliability in RNS arithmetic operations compared to existing solutions, with manageable overhead.