NRP

Projects

One full case study and a set of one-screen notes. Each entry carries its organisation, years, a result line, its stack and, where it applies, a mark: anonymised, projection, co-authorship. Figures are dated; where a number cannot be dated and sourced, it is not here.

Case studies

  • Origin–destination matrices from smart-card data for Bogotá's BRT and bus network

    About 3.8 million validations on a weekday (13 Feb 2026) turned into origin–destination matrices, load profiles and service indicators: ~68 % chained to a destination before expansion, 97–99 % after it, in about 15 minutes a day at near-zero marginal cost.

    TRANSMILENIO S.A., Dirección Técnica Zonal · 2024–2026 · R · SQL · GCP BigQuery · Parquet

Other work

  • A data platform for a three-person data team: Prefect, DuckDB and AWS

    Python and Prefect pipelines into a DuckDB-over-Parquet warehouse on AWS, serving embedded Power BI to 45 internal and about 30 external users; assistance-claims processing time cut by 70 % with renewed, AI-assisted processes.

    LifeOne (UnoGroup) · 2025–2026 · Python · Prefect · DuckDB · Parquet · AWS (ECS, Lambda, EC2, PostgreSQL) · Docker · Power BI

  • Risk data warehouse and versioned analytics apps at Colombia's central counterpartye

    A star-schema warehouse consolidating risk and market data for regulatory and internal reporting; the 'CRCC Data' platform of Shiny applications and APIs for internal teams and clearing members; automated risk back-testing and historical-VaR simulations.

    CRCC — Cámara de Riesgo Central de Contraparte de Colombia · 2022–2025 · SQL · Oracle · Jenkins · R · Shiny · Python

  • Six metaheuristics in Python, with statistical calibration

    Six algorithms, each on a classic problem, with parameters calibrated statistically rather than by hand; public repository.

    Universidad de los Andes, graduate course in metaheuristics · 2020 · Python

  • A multi-criteria MILP for intervention planning on ~1,500 national bridges

    A mixed-integer linear programme over about 1,500 national bridges that schedules interventions under a multi-period budget and integrates two criteria, instead of prioritising by a single score.

    Universidad de los Andes, with INVIAS (Colombia's national roads institute) and Universidad Javeriana · 2020 · Python · gurobipy

  • Credit-scoring models with XGBoost, SHAP and PSI stability checks

    Two credit-scoring models developed and documented for the risk team: a payroll-data scoring model with decision thresholds and an early-delinquency model, each with SHAP explanations and PSI stability checks.

    Banco de Bogotá · 2021–2022 · Python · XGBoost · SHAP · scikit-learn

  • A self-controlled case series on 55,114 HPV-vaccinated girls

    A population-based analysis of public data on 55,114 girls who received the first dose (39,488 the second), using a self-controlled case series design, which compares each person's risk within and outside the exposure window and so removes fixed between-person confounding.

    Universidad de los Andes, M.Sc. Industrial Engineering · 2022 · Conditional Poisson regression · Self-controlled case series

Marks

  • e co-authorship