Portfolio

Eight in-depth case studies. Eight decisions made clearer.

Each project below includes a downloadable one-page catalog you can share on WhatsApp, email or LinkedIn. Looking for the twelve editorial dashboards? See the analytics grid on the home page.

AI Data Analysis

Customer Segmentation — K-Means + PCA

E-commerce · 2,000 customers

Segmented 2,000 customers into four behavioural clusters using K-Means with PCA dimensionality reduction. Optimal K=4 via the Elbow Method and validated with a Silhouette Score of 0.255.

Key findings

• —Power Shoppers (n=617): 40 purchases/yr · $202 avg order.
• —Dormant Churners (n=282): 235 days inactive — retention risk.
• —Loyal Seniors (n=398): $79K income · 5.4-year membership.
• —Casual Browsers (n=703): still engaged, upsell opportunity.

Tells the business exactly which customers to retain, upsell, and win back — instead of one-size-fits-all marketing.

⬇ Download catalog (PDF)Discuss a similar project →

AI Data Analysis

RMS Titanic — Survival Analysis (Corrected)

Statistical modelling · 891 passengers

Four corrected analyses scoring 8.5/10. Fixed Class×Sex confounding, Age×Sex segmentation, MAR imputation, and CabinKnown collinearity most beginner analyses miss.

Key findings

• —1st-class men survived at 2.6× the rate of 3rd-class men.
• —Child survival advantage equal for boys & girls under 10 (~75%).
• —3rd-class missing Age 3–10× more — grouped imputation required.
• —CabinKnown adds <0.5% AUC once Pclass is included — redundant.

Demonstrates rigorous statistical thinking — not just running models, but knowing when results lie.

⬇ Download catalog (PDF)Discuss a similar project →

AI Data Analysis

Feature Engineering — Revised Analysis

Property pricing · 1,460 properties · 5-fold CV

Five production-grade fixes to a property pricing model. Final pipeline: R²=0.827, RMSE=$49K — all dollar metrics back-transformed for honest comparison.

Key findings

• —Mean-centred features: VIF 37→1 (multicollinearity resolved).
• —Smoothed encoding (k=10): rare neighbourhoods pulled to global mean.
• —Honest dollar RMSE: exposed log model as $1,501 worse, not better.
• —Ridge + StandardScaler pipeline: R² 0.485 → 0.826.

Shows how disciplined feature engineering turns a mediocre model into a deployable asset.

⬇ Download catalog (PDF)Discuss a similar project →

AI Data Analysis

Titanic Dataset — Exploratory Dashboard

EDA · 891 passengers

Full exploratory data analysis dashboard covering survival outcomes, age distribution, correlation structure, and class-sex survival interactions.

Key findings

• —Overall survival rate: 41.6% (520 perished, 371 survived).
• —Sex strongest predictor (Sex_male ↔ Survived = −0.51).
• —1st-class females: ~82% survival vs 3rd-class males: ~16%.
• —Engineered FamilySize feature from correlated variables.

A clean EDA baseline that surfaces relationships before any modelling decisions are made.

⬇ Download catalog (PDF)Discuss a similar project →

Data Analysis

Cloud Computing Dataset Catalog

Data cataloguing · Power BI style

Comparison of two cloud-computing data sources — one synthetic (Kaggle) and one real-world (MIT SuperCloud) — to guide dataset selection.

Key findings

• —Balanced sample: 1 synthetic / 1 real-world source.
• —Cloud Task Scheduling: best for supervised scheduling models.
• —MIT SuperCloud: best for realistic production-trace analysis.
• —Feature coverage matrix maps dataset → question.

Saves teams weeks by choosing the right dataset before writing a single line of code.

⬇ Download catalog (PDF)Discuss a similar project →

Data Analysis

Startup Dataset Analysis Dashboard

97,193+ companies · $300B+ funding

Comparative analysis of two startup data sources covering 97K+ companies and $300B+ in disclosed funding — built to evaluate providers for lead-gen and market-mapping.

Key findings

• —GrowthList covers the largest volume: 76,333 startups.
• —VCBacked: live-updated recent funding rounds and stage detail.
• —GrowthList: verified emails, social profiles and CSV export.
• —Source-selection KPIs prioritised over raw startup metrics.

Guides procurement decisions on third-party data — clear ROI on which source to subscribe to.

⬇ Download catalog (PDF)Discuss a similar project →

Data Analysis

Fintech Campaign Performance Dashboard

Banking · Q3 · 43,058 customers

Q3 campaign performance across 43,058 banking customers. Conversion rate 11.5% on 4,935 subscriptions, average call duration 248s, average balance €1,179.

Key findings

• —Longer calls increase subscriptions — quality > volume.
• —Higher balances convert better — segment by balance tier.
• —Seniors 65+ and students perform strongest — refocus targeting.
• —Over-contacting drops conversion — cap repeat attempts.

Four strategic actions: prioritise high-propensity segments, improve call quality, reduce repeat contacts, target prior-success profiles.

⬇ Download catalog (PDF)Discuss a similar project →

Data Analysis

ExcelGrow Industries — Sales & KPI Dashboard

Sales analytics · 2015–2017

Three-year sales dashboard covering $909K revenue, 33,798 units, and 9,971 transactions — regional, product, rep and seasonal breakdowns.

Key findings

• —Best region: West (44.9%). Best rep: Mike ($205.6K). Top product: Quad ($194K).
• —YoY growth: +1,687% (2016) then −1.15% (2017) — flag the plateau.
• —Seasonal Q4 spike in transactions — staff up accordingly.
• —Doublers has highest revenue per unit ($40.88) — protect margin.

Turns three years of sales noise into a one-page board view that drives next quarter's plan.

⬇ Download catalog (PDF)Discuss a similar project →