joshuavogel.dev

Case Studies

Technical post-mortems

Real engagements, real tradeoffs, real numbers. Each study covers the problem, the architecture decisions, the implementation, and the business outcome.

Series C B2B SaaS (confidential)

35%

Cutting BigQuery Spend 35% Without Slowing a Single Dashboard

BigQuery spend had grown 4× in 14 months to $96K/month while dashboard p95 latency crept past 12 seconds. Finance demanded cuts; the data team feared breaking the trust they had just rebuilt.

BigQuery dbt Airflow Looker Cloud Cost Reduction →

Mid-market E-commerce Platform

14 → 1

One Definition of Revenue: A Semantic Layer Finance Actually Signed Off On

Fourteen conflicting definitions of 'revenue' lived across Looker explores, dbt models, and spreadsheets. Board meetings opened with a 20-minute argument about whose number was right.

dbt MetricFlow Snowflake Looker Revenue definitions unified →

D2C Subscription Brand

99.95%

From Nightly Fires to Boring Reliability: Rebuilding a Kafka Ingest Pipeline

A hand-rolled Kafka consumer crashed silently several nights a week, duplicated events on every retry, and left marketing dashboards wrong by up to 8% — discovered only when spend decisions had already been made.

Kafka Snowflake Airflow Python Pipeline uptime (was 97.2%) →