withsoon

YouTube data platform — Start Here

01

Scope

Signals in

Products out

Interview Script

I’d design YouTube’s analytics and data platform. Playback, engagement, search, impression, ad, and backend events enter a durable Kafka backbone. Streaming jobs serve freshness-sensitive products such as live views, trending, QoE alerts, and online features. The same events land in an Iceberg lakehouse, where batch jobs validate, deduplicate, reconcile late data, and publish certified Silver and Gold products for Creator Studio, recommendations, experiments, monetization, BI, and backfills—with governance and quality around every layer.

02

Capabilities

01

Measure

Collect and qualify audience behavior

02

Decide

Turn fresh signals into product decisions

03

Operate

Serve creators, revenue, and trust

03

Platform Flow

YouTube signals

Fast path

seconds → minutes

Truth path

hourly → T+1
Governance · quality · lineage · security · metric versions · replay

“Playback and business events enter one replayable platform. Streaming serves freshness-sensitive products, while Iceberg and batch jobs publish certified data for creators, recommendations, monetization, experiments, and backfills.”

04

Scale Targets

Interview assumptions for sizing and trade-off discussion.

05

Correctness

FAST LANE Window + bounded dedupePROVISIONAL
CERTIFY LANE Full replay + quality gatesCERTIFIED
Live views · trending · alertscorrected byOfficial views · watch time · revenue