Melvin.

Four findings,
and the working.

Four questions I could not stop poking at. Each one written finding first, with the working underneath.

Read the firstAll projects

2017

01

the year the gender full-time gap started closing for a different reason.

Australian Labour Market Dashboard

End to end pipeline on live ABS data

Python ingests live ABS data, dbt models it into a tested star schema on SQL Server, and Power BI and Excel read the mart on top. Built to answer a question, not just move rows: the gap between male and female full-time work is closing, but the obvious explanation for why is wrong.

◆ Live ABS API

ABS API, refreshed on release  /  dbt star schema, 124 tests  /  Power BI and Excel, generated as code

Read the case study

38 hrs

02

is how long the average video survives on the Trending page before it vanishes.

YouTube Trending Analytics

Forensics on 40,000 trending videos

Forty thousand trending videos across ten regions: what actually predicts a spot on the board, how long a video holds it, and which signals turn out to be noise.

◆ YouTube Data API, 40k rows

YouTube Data API, 10 regions  /  pandas, then scikit-learn  /  Plotly notebooks

Read the case study

7.1%

03

parity rate in household, against 62.5% in pantry. The two chains compete where shoppers can price from memory and barely at all where they cannot.

Woolworths vs Coles Price Analytics

A year of prices, two retailers

Competitor benchmarking on public data: prices pulled daily from both retailers web APIs, matched into identical product pairs and tracked in a dbt warehouse. A year of history backfilled from an open tracker on the same product ids, checked against my own days at 99.97% before anything is built on it. Parity rate, mean absolute gap, promotion frequency and depth, store brand against name brand, and whether a price difference is a promotion or a position.

◆ Live retailer APIs, plus a backfilled year verified at 99.97%

Two retailer web APIs, 13 collected days, plus a backfilled year  /  rapidfuzz pairing, dbt with SCD2 history, pre-registered buckets  /  Parity, gap, promotion and repricing, by aisle and brand tier

Read the case study

37%

04

of the discount-promo cohort was still a customer at month six. Neighbouring cohorts kept about 71%.

SaaS Sales & Revenue Analytics

Cohort retention from 12.5K invoices

MRR, churn, NRR and CLV computed from 12,500 invoices through a tested dbt pipeline of eight models and forty-four data tests, with a cohort retention heatmap as the centrepiece.

◇ Synthetic, seeded generator

12,500 invoices  /  dbt, 8 models and 44 tests  /  Cohort retention heatmap

Read the case study