A lightweight machine learning pipeline designed for CubeSat image classification, optimizing data transmission efficiency in space. This project integrates preprocessing, model training, pruning, quantization, and evaluation to enable real-time decision-making onboard resource-constrained satellites. Inspired by the VERTECS mission, it ensures high accuracy while minimizing computational costs.
A machine learning model developed to predict credit risk and assign credit scores, supporting data-driven lending decisions for Bati Bank's Buy-Now-Pay-Later (BNPL) service in collaboration with an eCommerce platform.
This project leverages machine learning to detect fraudulent transactions in e-commerce and banking, aiding in proactive security and risk management. The goal is to provide a robust fraud detection pipeline with explainability, deployment, and dashboard visualization for actionable insights.
A comprehensive data warehouse solution for Ethiopian medical business data scraped from Telegram channels, including data scraping, object detection with YOLO, and ETL/ELT processes.
This project uses advanced time series forecasting models to enhance portfolio management for Guide Me in Finance (GMF) Investments. By analyzing historical data for Tesla (TSLA), Vanguard Total Bond Market ETF (BND), and S&P 500 ETF (SPY), we aim to forecast market trends, optimize asset allocation, and manage risk.
A comprehensive project analyzing the impact of significant political and economic events on Brent oil prices. This project includes data preprocessing scripts, statistical modeling using ARIMA and LSTM, exploratory data analysis (EDA) visualizations, and an interactive dashboard built with Flask and React. The goal is to provide actionable insights for investors, policymakers, and energy companies navigating the complexities of the oil market.
A machine learning solution to forecast sales for Rossmann Pharmaceuticals' stores across various cities six weeks in advance. Factors like promotions, competition, holidays, seasonality, and locality are considered for accurate predictions. The project structure is organized to support reproducible and scalable data processing, modeling, and visualization.
Building a real-time data ingestion and entity extraction pipeline for Amharic messages from Ethiopian e-commerce Telegram channels. Using fine-tuned LLMs, the system identifies key business entities like product names, prices, and locations, populating a centralized EthioMart platform. This project consolidates decentralized Telegram channels into a unified hub, addresses Amharic-specific linguistic features, and evaluates model performance for NER.
Built a Power BI report to analyze sales, profit, and performance trends across regions and categories by preparing and modeling the data, designing insightful visualizations, and managing report security.
Analyzed African population trends (1990–2023) using Power BI. Key insights include Nigeria's 15.4% population share, wide urbanization gaps, an average life expectancy of 65.4 years, Somalia’s highest fertility rate, and regional youth vs. elderly dependency imbalances.
Developed a dashboard to analyze water access data, highlighting issues like shared taps, well cleanliness, and wait times. Provided insights into gender equality and crimes against women, demonstrating the ability to turn complex data into actionable insights.
View more on GitHub.