English · Español
I am a final-year pure mathematics student at Universidad Central del Ecuador who works where rigorous mathematics meets real data. My training (real and functional analysis, measure theory, numerical analysis, PDEs, mathematical statistics) is what I use to check that an analysis is right, not only that it runs.
- Forecasting methodology for a national regulator. During my internship at Ecuador's hydrocarbon regulatory agency I authored a 73-page forecasting and anomaly-detection methodology (SARIMAX with intervention analysis, proposed to replace the 2017 one), with its Python pipeline and user manuals.
- Honest results. I audit pipelines, fix what is wrong and report what does not work: in a rain-prediction project I corrected four critical preprocessing errors and documented that the model did not beat a simple baseline.
- Verified numbers. I re-derive what libraries compute: in my survey-sampling project, 98 out of 98 estimates from R's
surveypackage were reproduced by hand.
| Need | What I do | Tools |
|---|---|---|
| Survey and official data | Estimation with sampling weights, strata and clusters; correct standard errors, design effects, domain estimates | R (survey, tidyverse) |
| Forecasting and anomaly detection | SARIMA/SARIMAX, intervention analysis, prediction intervals, monitoring rules | Python (statsmodels), R |
| Predictive modeling | Leakage-free features, temporal validation, calibration, comparison against baselines | Python (scikit-learn, XGBoost) |
| Reproducible reporting | Analyses where every figure is traceable to code; reports and documentation | Quarto, LaTeX, Word, Git |
| Mathematical content and AI training | Proofs, problem design and review, step-by-step solutions in Spanish and English | LaTeX |
|
How wrong is an analysis that ignores the sampling design? Ecuador's national student assessment (50,578 students): ignoring clustering understates standard errors 1.8 to 3.8 times; ignoring weights biases the mean by more than 10 points. R survey tidyverse
|
What must be fixed before trusting an official registry? Ecuador's 2021 death registry (107,648 records): 369,685 disguised missing values found and every cleaning decision justified. Python pandas SQL
|
|
Can a plane predict how deep Ecuador's earthquakes are? Existence and uniqueness of the minimizer proved; normal equations and gradient descent in NumPy, matching scikit-learn to 2.84e-13. Python NumPy optimization
|
What does a matrix do, geometrically? Linear maps, eigenvalues and SVD from scratch; image compression and an algebraic and numerical proof that PCA is the SVD of centered data. Python NumPy SVD
|
| Background (UCE coursework) | What it brings to data work |
|---|---|
| Measure theory I–II, probability theory | Knowing exactly what an estimator estimates and under which assumptions |
| Mathematical statistics, statistical inference, linear models | Sampling designs, likelihood, confidence intervals and tests done right |
| Linear algebra I–III | PCA, SVD, least squares and conditioning understood, not memorized |
| Real analysis I–V, functional analysis I–II, optimization | Convergence and optimization arguments behind machine learning |
| Numerical analysis I–IV | Stable computation: solving systems instead of inverting matrices, error control |
| PDEs I–IV, dynamical systems, mathematical modeling | Turning a physical or economic process into a model with explicit assumptions |
| Proof writing | Finding the hidden assumption that makes a result fail |
- Final year of the mathematics degree, taking Linear Models, Probability Theory, Optimization, Dynamical Systems, Mathematical Modeling.
- Adding bootstrap variance estimation and a Quarto report to the survey-sampling project.
- Open to part-time or freelance remote work in data science, statistics or math for AI training.
En español
Soy matemático (último semestre, Universidad Central del Ecuador) y trabajo donde la matemática rigurosa se encuentra con datos reales. En mis prácticas en la Agencia de Regulación y Control de Hidrocarburos elaboré una metodología de pronóstico y detección de anomalías basada en SARIMAX (propuesta para sustituir la de 2017); analizo datos en R y Python (encuestas con factor de expansión, series de tiempo, modelos predictivos) y verifico cada cifra antes de entregarla. Busco trabajo de medio tiempo o freelance remoto en ciencia de datos, estadística o entrenamiento de IA en matemática.
Portfolio From Mathematician to Data Scientist · it grows with every project