Data Scientist focused on statistical modeling, interpretable machine learning and auditable Generative AI.
I have a B.Sc. in Physics and I care about evidence: fair baselines, honest uncertainty, and knowing when a difference is real and when it is noise. The repositories below are personal projects built in my own time with public data.
| Project | What it shows |
|---|---|
| Pricing & Experimentation Lab | Price elasticity on 1M real transactions (DuckDB SQL, fixed effects, empirical-Bayes shrinkage), a bias found and corrected, and A/B-test sizing with CUPED |
| Colombian Electricity Demand Forecasting | SARIMAX, LightGBM and an ensemble on public XM data; 2-year rolling-origin backtest, Diebold-Mariano tests, conformal intervals, MLflow |
| Symbolic Regression Insights | PySR recovers Kepler's third law from NASA exoplanet data; readable formula vs gradient boosting on concrete strength |
| arXiv RAG Search | FastAPI hybrid-retrieval RAG over public arXiv abstracts with an evaluation harness, citation checks, local LLM, Docker and CI |
| Campaign Decisioning Lab | Leakage-aware propensity model that ranks outreach under a contact budget: calibration, lift, capture, bootstrap intervals |
| BriefSpec | Auditable GenAI triage with Amazon Bedrock structured outputs, strict validation and a synthetic offline evaluation |
Python · SQL · statsmodels · scikit-learn · LightGBM · PyTorch · PySR · pandas · Polars · DuckDB ·
Amazon Bedrock · SageMaker · MLflow · FastAPI · Docker · Git
- Five AWS certifications, including Machine Learning – Specialty, Machine Learning Engineer – Associate and Generative AI Developer – Professional.
- Coauthor of a peer-reviewed article in Journal of Physics A (DOI).
- Spanish (native), English (C1). Based in Bogotá, Colombia; open to remote data science roles.
Portfolio · LinkedIn · ORCID · Email
These are independent personal projects on open data. They contain no employer, client or academic code, data, prompts or business rules.