Selected work

Case studies

The decisions behind three independent projects, from analytics data models to browser-based interaction.

Independent analysis · Public sample data

Product Analytics Command Center

The decision

Could a product team trust the commerce funnel and revenue figures enough to prioritize a change?

What the project built

BigQuery SQL models turn nested GA4 events into sessions, cleaned purchases, product metrics and dashboard-ready views. Purchase events with usable transaction IDs are deduplicated per session. Records without a usable ID are flagged so their uncertainty stays visible.

What the analysis surfaced

The project documents 320 duplicate positive-revenue purchase rows removed and 9.26% of cleaned positive-revenue purchase records without a usable transaction ID. Intermittent cart tracking drove a later reporting cutoff for cart metrics. The published funnel counts sessions with each event and reports a separate sequence check.

Why a partner would refer this work

It shows how to trace an unreliable KPI to source events, document exactly what a dashboard measures, and give engineering a specific repair to make.

Figures are reported in the project’s analysis of Google’s public, obfuscated GA4 sample.

Independent project · Published survey data

The Shape of Happiness

The decision

How can readers compare life evaluations across countries while seeing gaps in the data and the limits of the underlying methodology?

What the project built

Santi Montouliu turned World Happiness Report data into a narrative and a linked explorer. A shared state store connects the D3 map, factor plots and country comparisons. Readers can share the current view and open data tables alongside several charts.

What readers can inspect

Compare up to three countries across 14 available reporting years from 2011–2025. The country history preserves gaps such as missing 2013 rankings. Scores shown for 2025 are averages from 2023–2025. The factor views use estimated contributions.

Why a partner would refer this work

It shows how a public dataset can become a navigable application with shared state, accessible alternatives, and methodology close to the relevant chart.

Independent project · Deterministic simulation

Overtone Sound & Waves Lab

The decision

How can a learner connect a change in harmonic content or phase to what they hear and what the waveform shows?

What the project built

Overtone synthesizes tones in Web Audio and computes the steady-state gain and phase of a six-band EQ. Separate files handle the model, audio lifecycle, rendering and presets. Guided experiments connect each control to an observable result.

What users can reproduce

Load the beats experiment to compare 220 Hz and 223 Hz tones. The readout shows a 3 Hz beat. Audio starts after a user action, and waveform and spectrum plots remain available with sound off.

Why a partner would refer this work

It shows how to build a responsive browser tool around a technical model, with visual feedback, guided workflows and explicit model assumptions.

Overtone applies a steady-state filter model to generated tones. The plots show the modeled signal.