Data Scientist · AI & Computer Vision · Data Engineering · BI & Analytics
Economist by training, data scientist in practice—7+ years building pipelines, models, and BI in banking, SaaS, and AI programs. I connect raw signals to models you can explain and views teams actually use, with economic reasoning when tradeoffs matter.
Structure messy data into stable signals—measured, documented, ready for the next step.
Estimate scenarios with models you can inspect; probability as a tool for judgment.
Turn analytical output into views teams actually use—fewer KPIs, clearer ownership.
Name tradeoffs and constraints before choosing—technical work connected to purpose.
Computer vision explored with honest scope—curiosity without overstated claims.
Most projects start by stabilizing messy inputs, then estimating with models you can inspect, then shipping interfaces people run week to week. When incentives or constraints shape the answer, I make that explicit before anyone treats a chart as policy.
Signal, model, decision
Computer Vision is an intentional growth lane, not a core claim yet; it is developed under the same decision standards used in Analysis, ML, BI, and Economics.
See roadmapProblem · Process · Impact
Educational open data on datos.gov.co was hard to extract and reuse at scale for ICFES analysis.
Connected the SODA API with Python (sodapy), loaded results into BigQuery, and built Data Studio views.
Reproducible ICFES 2019-2 pipeline—less manual extraction, faster exploration for decision support.
2016 plebiscite results were scattered across sources—hard to compare municipalities and departments consistently.
Python scraping workflow to aggregate territorial results and publish them in Google Data Studio.
Reproducible municipality-level vote intelligence for political and policy analysis.
Automation risk in Colombia needed local evidence beyond US-centric occupation probability studies.
Adapted Frey & Osborne-style modeling with World Bank STEP survey data and econometric adjustment in Python.
Policy-relevant read on skills, automation exposure, and sector-level labor market structure.
Demand-led notes
Three practical entry points shaped by questions already reaching this site.
Turn official IPC series into comparable peso values with an auditable Python calculation.
Connect tkinter controls to pygame playback, track state, duration, and volume.
Send generated files with modern Gmail authentication and run the job on a schedule.
A quoted loan rate describes nominal cash flows. It does not show how the payment burden changes after inflation. This note connects three views of the same ...
A calculadora de inflación Colombia is useful only when its source, reference months, and formula are visible. This implementation turns the official monthly...
Cloud chatbots are easy to demo and hard to trust with internal notes. RAG means Retrieval Augmented Generation: before the model answers, you fetch a few re...