Portfolio

JP Castro

Senior Data Architect & Engineer

25+ years turning messy ERP data into governed, query-ready assets. I design medallion architectures, Kimball dimensional models, and modern lakehouse platforms on Azure Databricks, dbt, Airflow, and PostgreSQL. Everything on this site is live, running in production, and built by me.

Get in Touch View Resume GitHub LinkedIn

What I Bring

I take messy ERP data and turn it into governed, query-ready assets. If your company runs on JD Edwards, SAP, Oracle, or any legacy system where critical data is locked in cryptic tables with Julian dates and five-character codes, I build the infrastructure that makes that data usable: clean schemas, automated pipelines, dimensional models, and live dashboards your team can actually trust.

Cloud Lakehouse

Azure Databricks, Delta Lake, Unity Catalog, Lakeflow DLT. Production pipelines that run in minutes, not hours.

Data Architecture

Kimball dimensional modeling, medallion architecture, SCD Type 2, data governance. Designed to last decades.

Full-Stack Delivery

I do not hand off specs. I build the pipeline, the API, the dashboard, and the deployment. One engineer, end to end.

● Live

JDE Data Platform

A full production data platform built on a simulated JD Edwards aerospace ERP. Data flows through a Bronze / Silver / Gold medallion architecture into live dashboards, status boards, a customer self-service portal, and an e-commerce storefront. The platform also includes an MDM layer that unifies customer records across five separate ERPs using Splink probabilistic record linkage, resolving duplicates like "Boeing Co." and "THE BOEING COMPANY" into golden records with consolidated sales visibility.

📖 Quick terminology (if you're new to modern data stacks)▾

ERP / JD Edwards

Enterprise software companies use to run operations — sales orders, inventory, purchasing, AR/AP. JD Edwards (JDE) is Oracle's ERP for mid-to-large manufacturers and distributors.

Medallion Architecture

A three-layer pattern for cleaning and organizing data: Bronze (raw copy from source), Silver (cleaned and typed), Gold (business-ready aggregations).

dbt (data build tool)

Open-source tool that transforms raw data into analytics-ready tables using SQL and version control. Runs the Silver and Gold layers.

Apache Airflow

Open-source workflow scheduler. Runs the full data pipeline on a nightly schedule and handles retries and alerting.

RFQ

Request For Quote — a customer asking for pricing on parts. Distributors deal with these constantly.

Railway

Cloud platform hosting everything you see — the Postgres database, the API, the dashboard, the portal, and the shop.

MDM (Master Data Management)

The practice of creating a single, trusted view of key business entities (customers, vendors, items) across multiple source systems that may store the same data differently.

Splink

Open-source Python library for probabilistic record linkage. Uses techniques like Jaro-Winkler similarity and Fellegi-Sunter models to match records that refer to the same entity but have different names, formats, or typos.