01
Problem
A live civic analytics platform showing how busy Melbourne's CBD is right now and forecasting the next 24 hours.
02
Data
- Ingests City of Melbourne open data from 99 pedestrian counting stations and 6,300+ in-ground parking sensors.
- An hourly GitHub Actions pipeline with a Pytest validation gate stores the data in Neon serverless Postgres.
03
Method
- A LightGBM model forecasts hourly foot traffic with no future leakage.
- Gemini Flash generates plain-English summaries.
- Runs entirely on free tiers ($0/month).
04
Results
The LightGBM model beats the 8-week historical baseline by 10% on 132,696 unseen sensor-hours.
