~/javier_
work.md

› cat about-my-work.md

I'm a data engineer. I build the pipelines that move data around and the small tools that make it useful for the people who have to decide something with it.

Open to data engineering opportunities. Client work below is anonymized: names, products and figures are synthetic, the engineering is real.

› ls ./projects

// swipe ↔

Retail analytics dashboard

live demo

// From a Power BI report to a phone-first web dashboard, fed by a nightly pipeline

Built for a small retail client, rebuilt here with synthetic data and a fictional store. Sales, products, customers, inventory and forecast on one page that loads fast on a phone.

./open live demo →
  • Read the client's Power BI model (5 pages, ~40 visuals, 60+ DAX measures) and rebuilt every number in Python to reconcile it.
  • Found that the original report overstated net sales by over a third: it counted quotes and added credit notes instead of subtracting them.
  • Nightly Airflow job pulls from the retail ERP API, writes parquet, and rebuilds a small aggregated data cube; a failed build never replaces the last good one.
  • ABC segmentation, a 4-week demand forecast valued in dollars, and stock-reorder rules (7 days of demand plus safety stock).
  • Hand-built SVG charts with no chart library, light and dark themes, and an (i) explanation on every chart.
  • Python
  • pandas
  • DuckDB
  • Airflow
  • parquet
  • vanilla JS + SVG
./open the live demo →

javuBot

open source

// A Slack control plane for pipelines running on AWS Step Functions

An open-source toolkit for data engineers who run pipelines on AWS: keep every run visible in Slack, start them from a slash command, and hold report publishing behind an approval. Planned as open source from the start, serverless, configurable and tested.

  • A Lambda listens to Step Functions status changes through EventBridge and posts the result to Slack: run time, the error on failures, who started it, and a link to the execution. It pings the on-call person when something breaks.
  • /javubot run <pipeline> starts a state machine from Slack. Pipelines are an explicit allowlist in a YAML file, and permissions are per Slack user ID instead of a shared password.
  • Approve before publishing: /javubot report approve or reject records the day's decision in DynamoDB, and a small gate Lambda lets state machines check it before they post anything.
  • Deployed with a SAM template; webhook URLs live in Secrets Manager. The command logic is independent of Slack, so it is covered by unit tests.
  • Python
  • AWS Lambda
  • Step Functions
  • EventBridge
  • DynamoDB
  • AWS SAM
  • Slack Bolt
  • Docker
./view the code on GitHub →

Self-hosted data platform

running 24/7

// A Raspberry Pi that runs real pipelines, a website and a media server

A home lab I run as if it were production: scheduled pipelines, a public site with no open inbound ports, private remote access and its own runbook.

  • Airflow runs the client ETL daily and publishes the dashboard data atomically after each run.
  • Public sites sit behind a Cloudflare Tunnel; gunicorn only listens on loopback, behind an unprivileged user and a hardened systemd unit.
  • Tailscale for private access, LAN-wide DNS ad blocking, and Jellyfin for family movie nights.
  • Everything documented in an operations runbook (setup, rollback, troubleshooting).
  • Raspberry Pi
  • Airflow
  • Cloudflare Tunnel
  • Tailscale
  • systemd
  • Docker

This website

you are here

// Django, a terminal-style dark theme and a Raspberry Pi

A small Django site I host myself: a bio, this projects page and the demos, all served from the Raspberry Pi above.

  • Django with SQLite, plain templates and a hand-written dark mono stylesheet (no front-end framework).
  • WhiteNoise with hashed static files, HSTS, secure cookies and a configurable admin URL kept behind Cloudflare Access.
  • Published through a Cloudflare Tunnel, so the home IP stays hidden and no inbound ports are open.
  • Django
  • SQLite
  • WhiteNoise
  • gunicorn
  • Cloudflare Tunnel
./get in touch →