Skip to main content

Case Studies

Work

Deep dives into selected projects — the data, the approach, and what the analysis actually showed

01/

My Spotify Listening, 2021–2026

Personal project · my own Spotify data

A personal-data pipeline from my own Spotify extended streaming history: 61,230 raw plays cleaned with pandas, genre-enriched via the Last.fm API, and published as a seven-view Tableau dashboard covering 4.7 years of listening.

TableauPythonpandasLast.fm API
02/

US Airline Flight Patterns

Coursework · IDS 405 Business Systems Analysis & Design

A four-point Tableau story across two years of US commercial flight records and 11 carriers: traffic over time, weekly flights by airline, and the distance-airtime relationship.

TableauTableau Public
03/

DocDefend+ — Grounding a Billing Agent in Real Terminology Data

Graduate coursework · UIC MS MIS · agent built for the Kaggle × Google 5-Day AI Agents Intensive

An ADK and Gemini agent that verifies every billing code against live NLM terminology data before it may enter a defensibility report, with a two-boundary PII firewall and a deterministic reconciler that overrides unbacked grounding claims. 16 of 16 evaluation cases pass, across 110 checks.

Google ADKGeminiMCPPythonClaude APIReact
04/

Parkinson's Voice Screening — Refusing an Age-Confounded AUC

Graduate coursework · IDS 506 Healthcare Information Management & Analytics, UIC MS MIS

Four classifier families by four feature sets on 81 sustained-vowel recordings. The best cross-validated AUC in the notebook is 0.912 — and the analysis exists to explain why that number cannot be claimed: age alone separates the groups at 0.864, and the defensible voice-only estimate is 0.775.

Pythonscikit-learnXGBoostParselmouthlibrosa