Built an LLM-based model that can solve puzzles presented in a modified version of the ARC-AGI dataset with about 32% accuracy.
I ranked #10 and was selected as one of the official winners of the Zindi-sponsored flights to Deep Learning Indaba 2025.
2024
Research
Wakanda AI: Engineering Smart Solutions for Mother Tongues by Advancing Language Technologies in the African Context
Bhiksha
Raj, Abraham
Nyete, Moise
Busogi, and
1 more author
In Africa, practical applications of speech technologies are scarce, often limited to adjusting pre-existing models. This project aims to overcome language barriers and enhance communication by developing innovative solutions that foster global understanding and inclusion. Key goals include establishing a data repository for several African languages, and developing speech and language models talored for local use cases.
A PyTorch re-implementation of RT-1 for the generation of actions represented as sequences of text tokens composed of the parameters to a CRAM generalized action plan.
Policy
Strategic Proposal for AI Implementation in Rwanda
A study on the use of ML4EO techniques in the development of end-to-end systems that can visually present the vulnerability
level of a selected regions in Rwanda (at the District or Province level).
2022
Course Project
RecSys challenge 2019: Session-based, Context-aware Recommender System in an Online Travel Domain
Cédric
Manouan, Benny
Uhoranishema, and Happiness
Karigirwa
In this study, we developed a session-based and context-aware recommender system to adapt a list of accommodations according to the needs of the user. The objective was to specifically forecast, based on the circumstances of each session, which hotels will trigger a click-out action from the user.
In this work, I take a look at the function that combines the low-level solutions in the Merge-Sort procedure and design my solution to the array merging problem similarly.
Course Project
Content-Based Recommender System for Scientific/Research Papers
Developed a Content-Based Recommender system for scientific papers as part of my personal project. The recommendation lists are generated using cosine similarity and text features extracted through TF-IDF.
Course Project
Remote Desktop Protocol (RDP): Flaws, Impact on African Businesses, and Mitigation Techniques
Deji
Adebayo, Arlette
Houndji, Comfort
Tenjier, and
5 more authors
This project outlines RDP, identifies its shortcomings, and assesses their implications for African businesses. The work aimed to offer practical suggestions for mitigating these issues, in order to foster a more conducive environment for sustainable economic growth and development in the region.
In this study, I explored a feature-based image stitching technique using different keypoints extraction methods and RANdom SAmple Consensus (RANSAC) algorithm to estimate the homography matrix between two input images taken from different viewpoints.
2020
Competition
Runmila AI Institute & minoHealth AI Labs Tuberculosis Classification via X-Rays Challenge
The purpose of this challenge was strictly educational, involving the exploration of machine learning and deep learning techniques for the development of a model capable of classifying tuberculosis and normal X-ray results.
Contributed to the development of a Pretrained Audio Neural Networks (PANN)-like model to classify audio utterances in Luganda (from Uganda) and English.
This hackathon was about predicting whether a person in an image is wearing a face mask or not. It was hosted by Zindi and lasted for 3 days (from Thursday 17 April to Sunday 19)
2019
Research
Implementation of an Intelligent System for Precision Agriculture: The Case of Cassava Production - ESATIC
The evolution of information and communication technologies (ICT) has enabled the development of various sectors. Consequently, the Ivorian government has initiated the Digital Solutions Project for Rural Area Access and E-Agriculture (PSNDEA) to create digital solutions that streamline work processes and enhance the productivity of rural farmers. The proposed solution is a system primarily based on two technologies: Artificial Intelligence (AI) and the Internet of Things (IoT). This system supports farmers in managing their resources to improve their overall performance. This work is divided into three parts. The first part introduces the hosting organization and outlines the project. The second part discusses the methodology employed. Finally, the third part covers the implementation of the solution and provides a discussion of the results.
Contributed to the research and development of the encryption algorithm used by an executable-only program. What makes our approach unique is its implementation in both Python and C++.
