Javier Moraga

Clinical reporting ran on legacy SAS infrastructure — fragmented across departments, with no lineage, no observability, and no path to real-time decision-making.

Led end-to-end migration to BigQuery with dbt-modelled semantic layer and automated CI/CD via GitHub Actions. Medallion architecture: raw → staging → marts. Tests on every mart before go-live.

85% reduction in query processing time. 5 legacy silos decommissioned. Real-time clinical analytics enabled for the first time. Deployed across Chile remotely.

Network disruptions at LATAM required live operational data. A 24-hour batch pipeline meant that every disruption response was based on yesterday's picture.

12M events/day from Jira and flight-ops systems · 5 countries · 40+ downstream dashboards · 120+ daily operational users.

Replaced the batch process with event-driven streaming on Pub/Sub + Dataflow + BigQuery. Exactly-once guarantees via Dataflow deduplication. Schema contracts enforced by Protobuf at ingestion; breaking changes blocked by CI.

Latency: 24h → 15 minutes. Network operations now steers live during disruptions instead of running next-day post-mortems. Distributed delivery across Chile, Brazil and Peru — async model throughout.

Data sources across the company had no formal documentation — access procedures, schemas, and ownership were tribal knowledge. New analysts needed weeks to become productive.

Created a comprehensive data source catalog covering every production data asset: user guides, access procedures, schema documentation, and source-to-report lineage maps.

Reduced information search time significantly across all analyst and BI users. The catalog became the standard onboarding resource for new hires joining the data team.

5M+ postpaid subscribers analyzed across billing cycles, commercial channels, payment methods, and geographic regions. Wireline + wireless product lines.

Diagnosed and fixed a RUT-K parsing bug in the AWK/sed notification pipeline that silently excluded ~24% of accounts. Rebuilt the extraction logic across 5 notification channels (IVR, SMS, APP, MAIL, WSP). Built risk quadrant scatter plots segmenting 16 regions by arrears rate vs. client volume — wireline and wireless separately.

Notification coverage: 76% → 99%. ~460K accounts recovered per billing cycle. Executive dashboards delivered to management with KPI tracking for suspension rates, notification reach, and commercial channel performance.

Rebuilt churn prediction model and underlying feature pipeline on multi-TB subscriber datasets using PySpark + Snowflake. Self-serve analytics dashboard deployed to commercial teams.

15% model accuracy improvement. 40% reduction in query operating cost. Churn interventions shifted from reactive to predictive.

Designed and operated end-to-end pipelines handling multi-TB network performance datasets. Automated manual reporting workflows across 8 business units. Introduced Kafka-based streaming for real-time network alerting.

40% elimination of manual effort. 99.9% data accuracy sustained. First real-time alerting infrastructure at the company.