Design Live Traffic and ETAs

Works out the report rate (150 million drivers, 480 reports each a day, is about 830,000 a second on average and 3.3 million at peak) and ingests it through stateless collectors into a log partitioned by region, so processing scales out and can fall behind briefly without losing data.

Noisy GPS points are matched to road segments using the sequence of a driver's points (an HMM or similar), in a stream processor keyed by driver or region, before any speed is computed.

Speeds are aggregated per segment over short sliding windows with an outlier-resistant statistic, blended with historical speeds when samples are few, and published every few seconds to a store keyed by segment that routing can read.

The routing engine absorbs new weights every minute without a full rebuild, for example customisable contraction hierarchies (metric-independent preprocessing plus a fast customisation step), and computes the arrival time along the route with live speeds near and predicted speeds further out.

Raw traces expire within 24 hours, only per-segment aggregates are kept, and a segment's speed is published only when enough distinct drivers contributed to it.

Every functional requirement in the brief is visibly served by something on the board, and the non-functional targets are addressed rather than ignored.

Components are labelled, data flows are drawn as connections between them, and the direction of each flow is unambiguous.