Local AI Upscaler — AMD GPU (Vulkan)
Single-file Python CLI that upscales AI-generated images for print on AMD GPUs in Windows · Real-ESRGAN NCNN + Vulkan, no CUDA, no cloud.
- Python
- Real-ESRGAN
- NCNN
- Vulkan
- +1 more
I have spent eight years as a geospatial data engineer and geoscientist. I built the first 3D geological model of the Coscuez emerald deposit and ArcGIS monitoring workflows for Colombia's maritime authority (DIMAR). At Arkho I shipped the AB InBev invoice pipeline that normalized 30 to 40 supplier formats with Textract and Step Functions, then the Casaideas SAP-to-cloud flow on Glue and PySpark. Each card below names the client, the stack, and the problem the work solved.
Single-file Python CLI that upscales AI-generated images for print on AMD GPUs in Windows · Real-ESRGAN NCNN + Vulkan, no CUDA, no cloud.
Claude Code skill that forces agents to find existing tools, libraries, and templates before reinventing the wheel.
Duplex fold-and-backlight QR puzzle generator. Print it, fold it, hold it to a lamp, and the hidden QR resolves.
A personal AI workflow system built on Claude Code, MCPs, and custom skills that operates across development, career management, and knowledge work.
Streamlit app that compares rent vs. buy housing scenarios for the Colombian market with ROI metrics.
Data engineering contributor on a large-scale logistics and distribution data pipeline for Coca-Cola Chile, with direct client interaction and quality-gated delivery.
Acting architect on a serverless AWS pipeline (Textract, Lambda, Step Functions) that normalized invoice PDFs from 30–40 supplier formats into structured data for downstream business workflows.
Built key stages of a Glue + PySpark pipeline that moves daily SAP extracts into a layered cloud warehouse, with lineage tying every record back to its source extract, for Latin American retailer Casaideas.
An open-source Python/Dash platform that monitors El Niño / La Niña (ENSO) conditions on Colombia's Pacific, built during my GIS work with the national maritime authority (DIMAR).
Built a geospatial COVID monitoring workflow for Colombia's national maritime authority using ArcGIS Online and Python, supporting institutional decision-making during the pandemic.
Owned the GIS information base and reporting cartography for Fura's Coscuez emerald operation in Boyacá, Colombia: QGIS and ArcGIS map series for investor, environmental, and licensing audiences, plus the first 3D geological model of the deposit and Python automation that cut ~100 hours of reporting work per year.