Developer based in Europe, working in the areas of Madrid and Amsterdam.

I enjoy a wide range of topics, including mathematics, AI/ML, theoretical concepts, especially in mathematics and computer science, finance, among others. I studied AI & Data Science at Maastricht University, Actuarial Science and Mathematical Finance at the University of Amsterdam (ongoing), and Sociology at the University of Granada.

Outside of formal coursework, I independently study topics that extend or complement my degrees. Recent subjects include category theory, measure theory, and information theory.

I also enjoy lifting heavy weights, rock climbing, running half marathons, diving, long-distance swimming, and generally engaging in hard (physical) activities.

• SphericalMesh.get_indices_at_coords: Implemented coordinate-to-index lookup for spherical meshes, converting Cartesian (x, y, z) to spherical coordinates and returning (r, θ, φ) bin indices (#3867, PR #3919).
• Model.description property: Added a description attribute to Model that serializes as a <description> element in the model XML, with round-trip support and automatic omission when empty (#3586, PR #3956).
• Material.get_activity chain parameter: Added an optional chain parameter to Material.get_activity() and Results.get_activity() that accepts a depletion chain for half-life values, falling back to the default ENDF/B-VIII.0 data for nuclides not in the chain (#3529, PR #3957).
• Collision count in particle splitting: Fixed CollisionFilter producing biased tallies when weight windows are active by preserving n_collision across Particle::split() instead of resetting to zero (#3916, PR #3958).

• reconstruct_from_patches_2d: Fixed incorrect image reconstruction when patch dimensions equal image dimensions (#10910, PR #33643). The overlap counting formula was missing a constraint on the number of patches.
• DecisionBoundaryDisplay in SVM example: Replaced manual contour plotting with DecisionBoundaryDisplay.from_estimator() in the SVM margins example (#33980, PR #34192), improving consistency with the modern scikit-learn API.

• lfilter early return on empty input: Added an early return path to scipy.signal.lfilter for zero-length input arrays, returning immediately with the unchanged initial conditions instead of passing empty data through the filter loop (PR #25334).

• L-BFGS-B optimizer respects disp=False: Fixed the optimizer printing convergence output unconditionally by passing the disp flag through to scipy.optimize.minimize (PR #9823).
• Remove dead cov_p assignment in GLM fit: Removed a redundant assignment to cov_p that was always overwritten by the hessian-based computation immediately after (PR #9826).
• plot_forest pass ax to dot_plot: Fixed CombineResults.plot_forest ignoring the caller's ax parameter, which created a new figure instead of reusing the provided axes (#8718, PR #9829).
• Fix GLMInfluence.hat_matrix_diag method name: Fixed a typo calling get_hat_matrix() instead of get_hat_matrix_diag(), which raised AttributeError when constructing GLMInfluence directly (#9415, PR #9830).

A study of 4,048,297 Pull Requests across 580 top open-source repositories (2016-2026). Covers contribution volume trends, temporal patterns, PR response times, contributor retention, the impact of AI coding tools via unsupervised changepoint detection, a composite project health index, and early warning signals for project decline. Data collected via the GitHub GraphQL API, analyzed with time-series decomposition, survival analysis, and forecasting models.

A menu bar app that captures all screens, runs OCR via Apple Vision framework, and highlights matching text in-place with a dimmed overlay. Real-time search, multi-monitor support, off-screen text detection via Accessibility API. Zero third-party dependencies.

Three-repo project to bring mathematical notation to the Xteink X4 e-reader. A firmware fork adds on-device MathML linearization to the CrossPoint reader firmware, rendering formulas from plain MathML elements without external dependencies. A QEMU-based emulator runs the real firmware binary on an emulated ESP32-C3 for testing without hardware. A CLI pipeline audits and adapts math EPUBs for the device: TeX-to-MathML conversion, XML repair, font coverage checks, and OCR-based PDF-to-EPUB with verification.