Economist → Data Scientist | PyTorch-first. Deep Learning · Bayesian Statistics · From-scratch, paper-faithful builds
Economist turned Data Scientist focused on Machine Learning, Bayesian Econometrics, Time-Series Analysis, and Graph Theory.
I apply advanced ML and econometric methods to analyze and forecast economic systems, from structural BVARs to graph-theoretic network models.
Always open to collaborating on impactful, data-driven research.
- From-scratch implementations: Faithful builds of core ML architectures from research papers — CNNs, RNNs, GANs/StyleGAN, YOLO — with clean modules, custom training loops, schedulers, and modern regularization (EMA, DiffAug, spectral norm).
- Econometrics in practice: BVAR & SGDLM pipelines for macro/finance; identification, uncertainty quantification, posterior IRFs/FEVD, and robustness/sensitivity analysis.
- Methodological rigor: combine DL with identification & uncertainty analysis.
Cross-domain ML/DL collaborations (vision, NLP, time series), open-source research tooling, and projects with clear social or policy impact.
|
|
| Project | Description | Tech |
|---|---|---|
| Bayesian SGDLM | Fully Bayesian SGDLM treating each node as a VAR(p) DLM; decouple–recouple filtering with Variational Bayes + importance sampling to learn sparse, time-varying cross-lag dependencies without inverting the full system. | Python Bayesian Time Series |
| Tourism ML Forecast | ML pipeline forecasting monthly foreign tourist arrivals in Colombian cities using Sentinel-2, economic, security, infrastructure and climate features; compares regression, tree-based and econometric baselines with KNN imputation, LIME and PDPs. | Python XGBoost Explainability |
| SBVAR-Col | Bayesian Structural VAR with agnostic identification to isolate U.S. Fed policy shocks and trace effects on Colombian macro-financial variables; Gibbs for reduced-form, MH for structural blocks, with IRFs and FEVD from posterior draws. | Statistics Python Bayesian VAR |
| Inflation Forecasting | Hybrid forecasting workflow combining ARIMA diagnostics (Stata) and LSTM tuning (Python) with dynamic forecasts and evaluation (MSE, MAE, R²). | Python TensorFlow ARIMA LSTM |
“Transforming data into high-impact decisions.”