I designed and shipped a production analytics platform on Google Cloud as the sole engineer — ingestion, orchestration, warehousing, monitoring, billing, and AI agents — for a national title-plant services company.
D. Bello ran its production reporting out of manual Excel files. I replaced that with the company's first cloud data platform: nightly Python ingestion on Cloud Run pulls 250+ county production sheets into BigQuery (batched at 25 tabs per request with exponential backoff), Airflow on Cloud Composer 3 consolidates hundreds of mismatched schemas by generating dynamic UNION ALL SQL from INFORMATION_SCHEMA, and the results feed dashboards, scheduled reports, Slack alerts, and a billing engine the business actually invoices from. Migrated from an early AWS design (S3, Glue, Athena, Lambda) to GCP, IAM configured end to end.
A Python billing engine that turns raw production tables into client-ready invoices — per-client pricing rules, split invoices, holds, hour-based models, and Excel/PDF package generation. Reconciled to the cent against historically issued invoices before going live.
A Next.js console on Vercel where leadership previews, reviews, approves, and sends every invoice — Google sign-in, a per-invoice review trail, and server-generated pages published to Cloud Storage on an hourly refresh.
Edit the pricing sheet and the console rebuilds itself: a sync DAG detects the change, publishes to Pub/Sub, and a single-instance Cloud Run dispatcher rebuilds and republishes the console — measured under four minutes, end to end.
Company-wide reporting rebuilt as BigQuery views with a Looker Studio dashboard on top, Monday-morning scheduled drops to Google Sheets with conditional formatting, and Slack report commands driven by a config sheet — no code change to add one.
The business's first in-flight-month revenue forecast: ratio-based projections computed from live production data, refreshed by a Cloud Run job, and streamed into an auto-updating dashboard.
A conversational agent over BigQuery so staff without SQL can ask production questions in plain language — Gemini function-calling against curated read-only tools, deliberately never text-to-SQL.
An hourly Cloud Run agent watching feed health, billing signals, and production pace — Slack alerts when a project falls 3+ counting days behind or a county's docs-per-hour moves 25% week over week. Tuned to stay quiet otherwise.
A multi-pass OCR engine for county land records: per-county indexing rulebooks, a human review queue, containerized batch runs on Docker, and accuracy scored pass-by-pass against ground truth — including an engine A/B evaluated on cost per document.
16 n8n workflows behind an MCP endpoint — turnaround-time reports, ingest triggers, invoice-send webhooks, report distribution — plus a scheduled agent that curates the team's knowledge base every evening from the day's actual work.
Ingest and reconciliation for third-party vendor billing files whose layout changes month to month — a dispatching parser handles each format, and per-client freeze guards keep issued invoices immutable once sent.
I'm an NCAA Division II cross country and track athlete at Biola University — 20+ hours a week of training on top of a full course load and a production engineering role. That's where the platform's personality comes from: show up every day, measure everything, trust the process, fix what the data says is broken.
I'm finishing a B.S. in Business Administration with an Analytics & AI concentration (expected May 2027), with a minor in Financial Planning — so I read a P&L as comfortably as a query plan. The systems I build aren't demos; a real business closes its books on them every month.