Data & Machine Learning Engineer — I build the data systems a real estate and media group in Mexico runs on: incremental ELT into Snowflake, anti-money-laundering regulatory reporting, geospatial dashboards, and machine learning on top of them. Actuary by training, which is mostly why I care whether a number can be defended.
Five of the projects below are interactive apps you can open right now. All of them run on synthetic data and need no credentials.
| Project | What it is | |
|---|---|---|
| Property Portfolio — Geospatial | Every property on one map. Areas measured geodesically from each parcel's own polygon; values restated for inflation. Built twice — R/Shiny and Python/Dash — over the same warehouse. | ▶ Demo |
| Land Valuation Model | What is this parcel worth, and how sure? Returns a calibrated range, not a number — 90% promised, 90.2% measured. Finds the purchases made outside the market. | ▶ Demo |
| Retail Space Manager | Leasing for three shopping centres. Releasing a unit never deletes anything — it posts negative square metres, so history and current state cannot drift apart. | ▶ Demo |
| AML Anomaly Detection | Unsupervised review queue for money-laundering risk, when nobody has ever labelled a case. 10.1× better than chance at a fixed review budget. | ▶ Demo |
| AML Regulatory Connector | Regulatory filings to Mexico's financial intelligence unit, end to end. The demo runs the production functions, unchanged, on invented contracts. | ▶ Demo |
| ERP → Snowflake ELT | Incremental load of 24 entities from a paginated ERP API. MERGE on a content hash that excludes the timestamps, because the ERP restamps rows it never changed. |
|
| Bank Reconciliation — Host-to-Host | Hourly reconciliation of a bank feed against ERP invoices, with a six-hour overlapping window because banks backfill. | |
| Bank Statement Consolidator | An Excel VBA macro rebuilt as an application: 78 classification rules, header-row detection, 525 movements, none unclassified. |
Python — pandas, scikit-learn, Streamlit, Dash, pytest · the pipelines, the
models and every live demo
SQL / Snowflake — incremental ELT, MERGE, data-quality views, Cortex agents
R — Shiny, sf, plumber · two production systems, and the reason some
repositories read as R-heavy
AWS — Cognito (OAuth2 + JWT), S3, CodeBuild
An AWS Machine Learning Engineer Associate certification, and moving the modeling work closer to production instead of next to it.
LinkedIn · pazjpaz17@gmail.com · soltecmty.com — independent consulting
📍 Mexico