Selectedwork.

// The problem

Legacy Alteryx and MySQL workflows couldn't keep pace with growing data volume. ETL ran for hours, gave teams no operational visibility, and required engineering to onboard every new data source by hand.

// My approach

Designed and built a cloud-native architecture on GCP: BigQuery for warehousing, Cloud Composer + Airflow for orchestration, Dataproc + PySpark for heavy transforms. The core innovation was a metadata-driven framework that lets analysts add new pipelines by editing config files — not writing code. Terraform managed everything as code; CI/CD shipped changes safely. Migrations ran incrementally so production never stopped.

// The problem

Engineering owned all operational visibility — logs, pipeline state, data checks, errors. Business teams had to file tickets to see what was happening with their own data. This created bottlenecks, slowed incident response, and eroded trust in the platform.

// My approach

Built a React + App Engine app that consolidates Airflow DAGs, BigQuery state, MySQL metadata, and pipeline logs into one operational interface. Real-time updates for status changes, structured search across logs and errors, one-click data validation, and controls to rerun, skip, or override pipelines without engineering tickets.

// The problem

Modeling client portfolios at scale meant training, retraining, and scoring hundreds of models in parallel. The old workflow lived in notebooks — slow, error-prone, manually triggered, and impossible to audit reliably.

// My approach

Built a metadata-driven ML platform on PySpark + Airflow. Data scientists configure new model runs via YAML; the system handles class imbalance correction, parallel training across client segments, model versioning, and automated scoring. Every run is fully reproducible from its logged config and metrics.

// The problem

ROI reports took 30 minutes to hours to generate manually and were prone to formatting errors. Quarterly business reviews (QBRs) ate entire days of slide assembly. Account managers spent more time formatting than analyzing.

// My approach

Built Django + React systems backed by BigQuery to generate ROI reports on demand (seconds) and assemble fully formatted QBR PowerPoint decks programmatically. The QBR engine reads from analytics tables and writes branded slides with consistent typography, charts, and narrative templates.