Live
📊 Analytics Dashboard
EdTech · Data Analytics
Student Performance Tracker
Upload a class result CSV and instantly see analytics — class average, score distribution, top performers, bottom performers, and students at risk. Gives teachers the data picture they have always needed.
Analytics · Pandas · Plotly
PythonPandasPlotlyStreamlit
Live
📚 Question Management
EdTech · Full-Stack · SQLite
CBT Question Bank Manager
Build, organise, tag, search, and reuse CBT question sets across subjects and exam years. One-click export to CBT Pro CSV format. No more lost or disorganised question banks before exam season.
SQLite · CRUD · Export
PythonSQLiteStreamlit
Live
⚠️ Early Warning
EdTech · ML · Classification · SHAP
Student At-Risk Predictor
Identifies students likely to fail before exam season — so teachers can intervene early, not after the results come back. Every prediction explained with SHAP — teachers see the reason, not just the risk score.
Random Forest · SHAP · Early Warning
PythonRandom ForestSHAPStreamlit
Live
📅 Study Planning
EdTech · Rule-Based · Personalisation
Student Study Plan Generator
Generates personalised study plans for students based on their exam targets, available time, weak subjects, and learning pace. Every student gets a plan built for them — not a generic timetable.
Rule-Based AI · Personalised
PythonStreamlitRule-Based
📊 Machine Learning Models
7 deployed models · 7 industries
Live
🏆 Highest AUC: 0.9393
🎓 3MTT Capstone
NLP · Binary Classification · Media
Fake News Detector — TruthLens Institute
Classifies news articles as real or fake using NLP — TF-IDF vectorisation, VADER sentiment analysis, and XGBoost. Trained on 12,999 real-world articles. The highest-performing model across all 12 projects. Formally assessed as 3MTT Capstone Project 2.
AUC 0.9393 · Accuracy 86.75%
PythonXGBoostNLTKTF-IDFVADERStreamlit
Live
📈 Only Regression Model
Regression · HR / Workplace Health
Employee Burnout Rate Predictor — NeuroWell Analytics
Predicts employee burnout rate as a continuous score (not just yes/no). 22,750 records, 108 GridSearchCV combinations, Gradient Boosting. R² of 0.855. Mental fatigue alone accounts for 85.9% of feature importance — a finding directly actionable for HR teams.
R² 0.855 · RMSE 0.072 · MAE 0.053
PythonGradient BoostingGridSearchCVStreamlit
Live
Binary Classification · Insurance
Insurance Claim Prediction
Predicts the likelihood of insurance claims from building structural features. 7,160 records. Random Forest + SMOTE to handle class imbalance + SHAP for explainability. Includes 6 business recommendations for risk-based pricing.
CV F1: 0.7921 · ROC-AUC: 0.6144
PythonRandom ForestSMOTESHAPStreamlit
Live
Binary Classification · HR / People Analytics
Staff Promotion Prediction — Yakub Trading Group
Replaces subjective promotion decisions with data-driven predictions. 38,312 Nigerian employee records — the largest dataset in the portfolio. Random Forest with class balancing. Removes favouritism and replaces it with evidence.
ROC-AUC 0.891 · Accuracy 93.6%
PythonRandom ForestClass WeightStreamlit
Live
Binary Classification · Banking / Finance
Bank Customer Churn Prediction
Predicts which bank customers are likely to leave — so the bank can intervene before they do. 10,000 records, Gradient Boosting. Key finding: age is the #1 churn driver, with a 7.43-year mean gap between churned and retained customers.
F1 0.6091 · ROC-AUC 0.8675
PythonGradient BoostingScikit-learnStreamlit
Live
Binary Classification · Social Policy / Finance
Income Level Prediction
Predicts whether an individual earns above or below $50K/year from demographic data. 48,842 US Census records — the largest dataset by record count in the portfolio. Random Forest + SMOTE + GridSearchCV. Best performer across 5 models compared.
Best ROC-AUC & F1 Across 5 Models
PythonRandom ForestSMOTEGridSearchCVStreamlit
Live
Multi-Class Classification · Logistics / Supply Chain
SwiftChain Delivery Delay Prediction
Predicts delivery delay categories (on-time, slightly late, severely late) for a logistics company. 15,549 orders, Gradient Boosting. Key insight: Standard Class shipping has a 38.1% late rate and accounts for 67.2% of model importance — actionable for operations teams.
Weighted F1 0.5791 · CV F1 0.5768
PythonGradient BoostingMulti-classStreamlit