This repository contains three compact R projects demonstrating dashboarding, AI-assisted reporting, and interpretable classification. It is an earlier implementation portfolio and complements the newer decision-first case studies linked below.
Historical portfolio note: This repository preserves the original May 2025 scripts and Excel outputs as evidence of an earlier stage of my AI-augmented analytics work. The main R scripts were revised in September 2026 after later review exposed analytical and implementation weaknesses. Both versions are visible so the development can be inspected directly; newer repositories demonstrate my current methodology.
| Project | Original 2025 implementation | Revised 2026 implementation |
|---|---|---|
| GPT Reporting | Original script · Original workbook | Revised script · evidence supplied to every model request, safer claims, configurable model, API error handling |
| ML Modeling | Original script · Original workbook | Revised script · explicit positive class, corrected metric semantics, stronger preprocessing and limitations |
The historical workbooks remain paired with the original scripts. Revised workbooks are intentionally not committed, because this repository documents the change in method rather than replacing its historical evidence.
| Project | Purpose | Stack | Evidence |
|---|---|---|---|
| System Performance Dashboard | Explore regional reliability, defect-rate, and downtime KPIs | R, Shiny, tidyverse, ggplot2, DT | Runnable app, sample data, screenshots |
| Marketing Report Generator | Produce a structured Excel report with data-grounded AI narratives | R, OpenAI API, dplyr, openxlsx | Script, sample data, generated workbook |
| Customer Response Classifier | Compare logistic regression and random forest on a held-out test set | R, tidymodels, ranger, yardstick | Script, sample data, evaluation workbook |
- Reproducible data preparation in R
- Interactive Shiny dashboards
- Structured Excel publication
- API-based narrative generation
- Classification workflows with held-out evaluation
- Transparent limitations and reusable sample data
These are focused demonstrations rather than production systems. The modeling project is a baseline comparison without hyperparameter tuning or cross-validation. The AI-reporting project requires a valid API key and sends summarized sample data to the configured model.