33 projects and systems built by Dominion Akinrotimi. From AI waste management for Nigeria to clinical ML research and open-source data quality tools.
Showing 43 of 43 projects
An open-source tool that checks a CSV or Excel file and says, in plain language, whether it is safe to use and what changed since the last one, with the exact rows behind every finding. Runs on your own computer: no upload, no account.
Highlights:
• Desktop app, command line and Python library
• Research preprint live on Zenodo (DOI 10.5281/zenodo.22211750)
• Free lessons on the maths behind it
Reads a bank statement in the browser and shows real income, spending and savings, with every row checked against the balance. The statement is never uploaded.
Highlights:
• Every transaction typed; nothing guessed
• Row-by-row reconciliation catches missing months
• Shareable money cards and proof-of-income reports
The loan origination data system for Cent: from application to compliance archive, with personal data encrypted field by field.
Highlights:
• AES-256 encryption on 16 PII fields
• 5-stage loan workflow as a state machine
• Dual sync: Supabase/PostgreSQL and Google Sheets
Led end-to-end delivery of a financial anti-corruption and transparency platform for an Ondo State Government agency, as Managing Director of COMBUF LTD.
Highlights:
• Owned decisions, developer guidance and stakeholder reviews
• On-site reviews and final handover to the client
Applied analytics and machine learning for clients in FinTech, cybersecurity, agriculture, manufacturing and public health.
Highlights:
• 26.1M addressable FinTech market identified
• 40% faster cyber-threat identification; 25% better rice-yield forecasts
• 30% safety non-compliance found; obesity study pretest (n=30, α=0.77)
A comprehensive strategic data analysis identifying Nigeria as the most compelling voice banking opportunity in Africa. Processed World Bank data to quantify a 24.4% addressable market gap.
Highlights:
• Analyzed 46.2 million adult market for voice banking
• Processed World Bank Global Findex Database (2011-2024)
• Built actionable dashboards & strategic metrics
Machine learning framework for personalized survival prediction for breast cancer patients using clinical data.
Highlights:
• METABRIC cohort: 1,981 patients with known outcomes, 21 clinical features
• Random Survival Forest: C-index 0.686 (5-fold CV), 0.691 on held-out patients
• Risk groups: deaths from 32.5% (low) to 82.6% (high)
Classifies H&E breast histology images as normal, benign, in situ or invasive carcinoma: 85.0% accuracy on 80 held-out images, with exact visual explanations and a live app.
Highlights:
• 85.0% accuracy on 80 held-out BACH images; every carcinoma in the test set detected
• 20 model designs compared fairly: image resolution mattered more than network size
• Exact explanations: activation maps and a SHAP value for each image patch
• Live app that refuses images that aren't histology
Leveraging AI techniques like pre-trained generative models and transfer learning, fine-tuning StyleGAN2-ADA and BigGAN on wildlife datasets to generate images of endangered species.
Highlights:
• Published on ResearchGate (DOI: 10.13140/RG.2.2.22698.63686)
• Fine-tuned StyleGAN2-ADA and BigGAN
• Addresses data scarcity for endangered species
Comprehensive ML system for online banking fraud detection with economic impact analysis for financial institutions.
Highlights:
• XGBoost model: AUC-ROC 0.8730
• Reduced false positive rate from 99.45% to 7.87%
• Demonstrated potential for significant cost savings in banking
Advanced ML and DL models to predict how changes in national and international policies affect global refugee movements, analyzing UN datasets.
Highlights:
• Analyzed UN datasets for refugee statistics and policy indicators
• Time-series forecasting for displacement patterns
• Data-driven insights for policymakers and humanitarian agencies
Collaborated with Insight Innovators on African food system challenges (production efficiency, supply chain economics, and waste reduction) across North, West, East, and South Africa.
Highlights:
• Analyzed food system data across four African regions
• Quantified economic impacts of food loss
• Developed policy recommendations
Interactive dashboard analyzing diabetes risk factors from public health data for preventive healthcare initiatives.
Highlights:
• Processed 10,000+ records with 15+ risk factors
• Identified 3 key factors accounting for 68% variance
• Reduced analysis time by 50%
AI-powered waste management for cities. Uses predictive analytics to forecast waste generation patterns using traffic, population density, rainfall, and day-of-week data | no expensive sensors needed.
Highlights:
• 99.3% accuracy forecasting collection needs
• Voice assistant in Pidgin, Yoruba, Hausa, Igbo
• Route optimization: up to 30% fuel cost reduction
• Online learning | system gets smarter with every collection
Full-stack platform enabling freelancers, consultants, and businesses to collect, manage, and showcase client testimonials professionally with centralized feedback and customizable categories.
Highlights:
• Three auth methods: Google OAuth, Magic Links, Secret Keys
• Custom review categories and profile pages
• Email notifications & management dashboard
• Rate limiting, security headers, API versioning
Platform dedicated to unlocking human potential beyond the resume. Connects individuals, employers, and institutions through a skills-first approach aligned with real strengths.
Highlights:
• AI-powered assessments for strength-aligned careers
• Personalized career roadmaps & dynamic portfolios
• Employer talent matching based on actual skills
• AI Career Coach & Resume Agent (coming soon)
Privacy-first, client-side AI system transforming body measurements into actionable style guidance. Addresses the $128 billion annual cost of apparel returns due to poor fit.
Highlights:
• Probabilistic body typing (Hourglass, Pear, Apple, etc.)
• Instant size conversion (US/UK/EU/AU)
• Personalized style recommendations
• AI image body scan & 3D virtual try-on
Formula-driven attendance tracking in Google Sheets with cryptographic hash verification and circular reference protections. No additional software required.
Highlights:
• Self-validating checksums using built-in formulas
• Protection against retrospective data alteration
• Published guide on Medium
• Designed for resource-constrained schools
Analysis of phishing vulnerability via questionnaire study in Nigeria, identifying key factors for cybersecurity awareness training.
Status: Completed
Tech Stack: Power BI, Data Viz, Cybersecurity
View DetailsExplored sales dataset for top-selling products, geographical insights (San Francisco dominance), and monthly/quarterly trends.
Status: Completed
Tech Stack: Power BI, Data Modeling, DAX
View DetailsEarly, on-time or late prediction for 15,549 orders, rebuilt without test leakage. Macro F1 0.48 against 0.24 for the baseline; 87% precision on late alerts.
Status: Completed
Tech Stack: Python, scikit-learn, Gradient Boosting
View DetailsGradient Boosting model predicting employee burn rate from 21,626 HR records. R² 0.905 on held-out validation data, explained with SHAP and LIME.
Status: Completed
Tech Stack: Gradient Boosting, SHAP, LIME, Streamlit
View DetailsTF-IDF text classifier on 6,306 articles. Tuned Logistic Regression reaches 92.2% accuracy on 1,262 held-out articles.
Status: Completed
Tech Stack: NLP, TF-IDF, scikit-learn
View DetailsLogistic regression model detecting offensive content in Yorùbá, a low-resource African language. 91.3% accuracy.
Status: Completed
Tech Stack: NLP, Logistic Regression, TF-IDF
View DetailsPython-based language detection using Multinomial Naive Bayes and TF-IDF with a user-friendly GUI for real-time prediction.
Status: Completed
Tech Stack: TF-IDF, Naive Bayes, Python, Tkinter
View DetailsInteractive language learning tool with voice interaction providing word definitions, examples, and pronunciation assistance.
Status: Completed
Tech Stack: NLTK, Google Speech API, Python
View DetailsComparative performance analysis of Dijkstra's, Bellman-Ford, and Floyd-Warshall on synthetic and real-world graph datasets.
Status: Completed
Tech Stack: Python, NetworkX, Matplotlib
View Details1st-place SDG hackathon project (SDG 4: Quality Education). Developed LearnVerse educative app with a team of five.
Status: Hackathon Winner
Tech Stack: App Dev, SDG 4, Team Project
View DetailsC++ guessing game where the user has eight tries to guess a randomly generated secret number between 1 and 100.
Status: Completed
Tech Stack: C++, Game Logic
View DetailsExamination of bike purchases using customer demographics and purchasing data with interactive Excel dashboard visualizations.
Status: Completed
Tech Stack: Excel, Dashboard, Data Analysis
View DetailsAnalysis of Java June coffee shop locations across Europe, exploring review counts, ratings, and business expansion insights.
Status: Completed
Tech Stack: Data Analysis, Visualization, Statistics
View Details3-month capstone analyzing sales data; identified top regions/countries by profit and provided management advice via Pivot tables.
Status: Completed
Tech Stack: Excel, Pivot Tables, Charts
View DetailsThree interactive science quizzes (Astronomy, Computer Literacy, Basic Physics) with shuffled questions, feedback, and retakes.
Status: Completed
Tech Stack: Python, Jupyter Notebook, CLI
View DetailsA collection of visually compelling design works created alongside data science pursuits.
Status: Portfolio
Tech Stack: Graphic Design, Visual Design
View DetailsContent-based recommender over 28,932 TMDB films blending plot TF-IDF with director and cast. 0.333 held-out genre overlap against 0.154 for random picks.
Status: Completed
Tech Stack: Python, TF-IDF, scikit-learn
View DetailsFour regression models on 278,860 policies showed the features cannot predict premiums (R² ≈ 0), and set out the data that would.
Status: Completed
Tech Stack: Python, scikit-learn, Ridge, Random Forest
View DetailsA/B test on 518,565 users: ads lifted conversion from 2.25% to 2.81% (z = 4.28), with a frequency-cap test recommended next.
Status: Completed
Tech Stack: Python, statsmodels, A/B testing
View DetailsK-means on 700 smartphone users finds five usage levels; ANOVA and chi-square tests show age has no significant effect.
Status: Completed
Tech Stack: Python, K-means, DBSCAN, SciPy
View Details17 business questions answered in PostgreSQL on a 15-table rental database: customers, stock, actors and store profit.
Status: Completed
Tech Stack: SQL, PostgreSQL
View DetailsStreamlit app estimating diabetes risk from eight health measurements. Logistic regression on 768 patients, 75.3% accuracy on held-out data.
Status: Completed
Tech Stack: Python, scikit-learn, Streamlit
View DetailsExcel analysis and slicer dashboard of 10,281 supermarket customers: 72% in the USA, 55% on bronze cards, February the busiest month for sign-ups.
Status: Completed
Tech Stack: Excel, Pivot tables, Dashboards
View DetailsProfiling and cleaning 5,850 movie and TV titles with pandas, treating missing values by what they mean.
Status: Completed
Tech Stack: Python, pandas
View DetailsLeak-free next-day temperature and rain forecasts for Lagos, tested on 2026. Beats "tomorrow = today" by 16 to 18% on temperature.
Status: Completed
Tech Stack: Python, scikit-learn, Time series
View DetailsA comprehensive overview of the tools, languages, and frameworks I use to build solutions.
I build data pipelines, machine learning systems and analytics that hold up in real use. Tell me what you are working on.