Commerce + Risk Analytics (DS → DE → AI-ready)

Portfolio-grade operational analytics product built from:

1. Olist e-commerce delivery intelligence: predict late deliveries and identify actionable drivers.
2. PaySim finance add-on (optional): fraud classification + “risk” marts (Week 3+).

Why this exists (recruiter framing)

• Business problem: late deliveries drive dissatisfaction and support cost.
• Analytics output: operational slices + executive-ready visuals + a simple model for triage.
• Engineering output: reproducible pipeline that exports dashboard-ready datasets and reports.

Repo map (where to look)

• notebooks/01_eda_olist.ipynb: Week 1 analysis notebook (learning + interpretation)
• notebooks/02_portfolio_story.ipynb: Week 2 presentation notebook (executive flow)
• src/pipelines/week2.py: Week 2 reproducible run (exports)
• reports/executive_summary.md: 1-page executive output (generated)
• docs/dashboard_recommendations.md: Tableau Public dashboard spec
• dashboards/tableau/screenshots/: generated dashboard screenshots for GitHub and LinkedIn

Architecture (high level)

raw CSV → processed Parquet → modeling dataset → models + slices → exports + executive summary + dashboard extract

Top operational findings

• Overall late-delivery rate is 7.9%, with an on-time rate of 92.1%.
• Late orders are delayed by 9.6 days on average, making late delivery a customer-experience and support-risk problem.
• Higher-value orders and high freight-ratio orders show elevated late-delivery risk.
• Highest customer-state late-delivery rates include AL, MA, PI, CE, and SE.
• The model is best positioned as an operational triage signal, not an automated decision system.

Quick start

python -m venv .venv
source .venv/Scripts/activate  # Windows (Git Bash)
# source .venv/bin/activate    # macOS/Linux
pip install -r requirements.txt

Data (manual download)

Place raw files here (not committed):

• data/raw/olist/ (all Olist CSVs)
• data/raw/paysim/ (PaySim CSV; optional for Week 2)

Ingest:

python -m src.ingest.olist
python -m src.ingest.paysim

Week 2: run the portfolio pipeline

This generates metrics, tables, figures, dashboard extracts, and an executive summary.

./scripts/run_week2.sh

Outputs:

• reports/metrics/late_delivery_model_metrics.csv
• reports/tables/late_delivery_by_customer_state.csv
• reports/figures/late_delivery_by_price_band.png
• reports/exports/olist_dashboard_extract.parquet (dashboard contract)
• reports/exports/model_metrics.csv (Tableau Model Insights helper)
• reports/exports/feature_importance_top15.csv (Tableau tooltip/helper)
• reports/tableau/*.csv (chart-ready Tableau sheets)
• reports/executive_summary.md

Tableau Public dashboard (Week 2)

• Published dashboard: E-Commerce Delivery Risk Intelligence
• Spec: docs/dashboard_recommendations.md
• Automation strategy: docs/tableau_automation_strategy.md
• Build guide: dashboards/tableau/BUILD_GUIDE.md
• Data: reports/exports/olist_dashboard_extract.parquet
• Faster Desktop start: ./scripts/open_tableau_package.sh
• Portfolio assets: docs/portfolio_launch_assets.md

Screenshots

Resume-ready proof

• Built a reproducible Python analytics pipeline for 99K+ e-commerce orders, generating executive KPIs, operational risk slices, model metrics, Tableau-ready extracts, and portfolio screenshots.
• Published a Tableau Public dashboard showing late-delivery risk by price band, freight ratio, weekday, customer state, and customer-seller lanes.
• Framed ML output as decision support by comparing baseline and random forest models, documenting leakage guardrails, operational limitations, and business recommendations.