I build spatial data pipelines, GIS systems, and AWS data infrastructure. 8+ years turning domain-heavy, ambiguous problems into systems that ship, across geosciences, geospatial engineering, and AI-assisted automation. Remote-first from Bogotá.
I still think in formations and ages, so my career stacks like a stratigraphic column: youngest at the surface, geoscience basement at depth. Hover or focus a bed to read the log.
A thin surface layer, still being deposited. Side-practice AI tooling I build on the Claude stack, sitting on top of the production work below.
Well-cemented and current. Geoscience software at Imaged Reality, where I sit between enterprise oil and gas clients, the product, and the Python that ships.
Porous and permeable, the reservoir of the section. Thick, well-bedded AWS pipelines at Arkho that moved retail, beverage, and document data at production scale.
Finely laminated, one map bedded over the next. Operational GIS and dashboards at DIMAR and Heinsohn that institutions and ministries put to work.
Crystalline basement, deformed and intruded, the foundation under everything above. Field geology, stratigraphy, and emerald resource modeling at Fura, where the domain intuition formed.
I built the first 3D geological model of the Coscuez emerald deposit, a geospatial COVID-monitoring workflow for Colombia's national maritime authority, and a serverless AWS pipeline that read invoice PDFs across 30–40 supplier formats for AB InBev. Each card below names the client, my role, the stack, and what shipped.
Colombia's national maritime authority (DIMAR) needed live operational status across its sites during the pandemic. I designed the workflow end to end: Survey123 forms in the field, GeoPandas cleaning and georeferencing the records, and ArcGIS Online dashboards that institutional teams read for daily planning.
No two suppliers laid out their invoices the same way, so accounts-payable read 30–40 PDF formats by hand. As acting architect I built a serverless AWS pipeline (Textract for extraction, Lambda for parsing, Step Functions for orchestration) that turned those PDFs into one structured, queryable schema.
Built the first 3D geological model of the Coscuez emerald deposit and owned the GIS reporting base behind investor, environmental, and mining-licensing maps across the emerald belt. Python automation of the repeated map and report production cut roughly 100 hours of manual work off the team's year.
I'm a geoscientist who became a geospatial data engineer. I studied at Universidad de los Andes, logged core in the field, then built the first 3D geological model of the Coscuez emerald deposit. From there I moved into Python, GIS, and spatial data: QGIS, GeoPandas, PostGIS, GDAL, raster and vector. Across 8+ years I build geospatial and spatial-ETL pipelines, GIS systems institutions keep in daily use, and AWS data infrastructure on Lambda, Glue, Step Functions, and Textract: 15+ ETLs across 4 CDK projects at Arkho.
I take domain-heavy, half-specified problems and ship working systems. As acting architect I built a serverless invoice pipeline that normalized 30–40 supplier formats with Textract, Lambda, and Step Functions, shipped COVID geospatial monitoring for Colombia's maritime authority, and built the Random Forest facies model behind an SPE paper at ADIPEC 2025. At Fura, Python automation of the reporting maps saved about 100 hours a year. I scope it and write the code, then state the tradeoffs so a stakeholder can act on them.
Field notes on geospatial tooling and the Claude Code workflows I build to make messy work repeatable.
I lean on Obsidian, Claude Code, Astro, and daily Python on Windows. Which ones earned a permanent slot, and which ones I dropped.
I took a pile of inconsistently named repos and gave them a clean <scope>-<what> convention. Then I set a README pattern and a deliberate public/private split.
How I wire Claude Code, custom skills, and Model Context Protocol servers into development and knowledge work. Small automations stack into a system I lean on every day.
I build geospatial systems, GIS and ArcGIS work, and AWS data pipelines, and I work in Python every day. Send me the messy version of the problem and I'll tell you whether it's a fit and what I can ship in the first week.