AI · NLP · STRUCTURE
Systems that read structure.
Building machine learning on messy data — Inka khipus, Southern Quechua, computer vision — and operations systems for LATAM teams.
Scientific poster, EEML 2026 Cetinje (fully funded) · arXiv:2607.00185
Work selected
06 projects
Why this work
Khipus
The Inka recorded administration in knots. Structure is the data. That is why this is ML on the Open Khipu Repository, not another chatbot.
Quechua
Most tokenizers assume spaces and European morphology. Southern Quechua does not. If NLP is going to work in the Andes, the tokenizer has to.
LATAM
I ship from Lima for teams that still run on WhatsApp, sheets, and handoffs. The person who reads the paper also automates the invoice.
I started in UX and frontend.
I now build ML pipelines, low-resource NLP, and agentic workflows.
Presented Inka khipus at EEML 2026. Writing the next paper on khipus as graphs.
Flows by Maky is the practice: I design, code, and ship the system myself.