Analytics engineer and data lead based in Santiago, Chile. I build the data platforms that subscription businesses run on: reliable pipelines, dbt models with tests, one definition per metric, and AI tools that answer from those definitions instead of guessing.
Today: Data & Analytics Manager at a multi-brand subscription sports-streaming platform. I built the data function from zero and own it end to end: Python ELT on AWS, Redshift, dbt, monitoring and alerting, BI, forecasting and churn modelling.
An end-to-end analytics engineering project for a fictional sports-streaming subscription app, built on synthetic data so it can be public. It runs locally on DuckDB and targets Snowflake.
- Layered dbt project (staging → intermediate → marts) with model contracts, unit tests and reconciliation tests
- Subscription finance: MRR with a full movement bridge, SCD2 prices, constant-FX reporting
- Web analytics: GA4 export events, sessionization, identity stitching and attribution
- Semantic layer in MetricFlow, reconciled against the marts in CI
- Orchestration with Dagster: dbt models as assets, tests as asset checks, freshness policies
- AI analyst over governed metrics: an MCP server with no raw SQL and a
verify_numbergate that re-derives every figure - Experiment readout: sample-ratio check, power, confidence intervals and CUPED
- CI on every pull request: linting, build and tests, impact report and opt-in AI review
SQL dbt Python AWS (Lambda, Step Functions, EventBridge, S3) Redshift BigQuery Snowflake DuckDB Dagster MetricFlow GA4 QuickSight Power BI Claude Code MCP
- One definition per metric, tested and reconciled before anyone reads the number.
- Every change goes through a pull request with CI; merge equals deploy.
- AI-first in daily work, with a human (or a second query) checking every figure.
LinkedIn · Spanish (native) · English (C1/C2) · German (B1)
