Germán PaduaMathematics · Engineering · Data science
Visual theme
ES — Español
Sections

Germán Padua Almería, Spain

AI, mathematics, data and curiosity.

Mathematician, computer scientist and data scientist. I work on credit risk models and build AI and machine learning projects.

See the projectsExperience
Photo · QGCN 2D projection

An idea brought to code: geometry and graph neural networks.

Projects

Selected research, engineering and applications.

More projects
  • Almería Isometric

    An isometric map of Almería built from real public data —PNOA LiDAR, orthophoto and Cadastre— where AI only decides what surfaces look like. Geometry is imposed by the data, not by the model.

  • Mi Balance

    A PWA for iPhone that records income and expenses and calculates the monthly savings rate. The data never leaves the device, backups are encrypted and it works offline.

  • inference-obs

    A self-hosted, observable LLM inference gateway. One operator serves several projects with virtual keys, per-project budget and every call traced down to the project and the end user who made it.

  • F1 Data App

    A published web application that analyses Formula 1 races and qualifying sessions from the public Ergast and FastF1 APIs: the grid, lap times, position evolution and pace distribution.

  • Football analysis with computer vision

    Player, referee and ball detection on match video with YOLOv8, team identification, trajectory tracking with distance and speed, and position heat maps.

  • Floor plan and renovation editor in 3D

    Apartment editor that runs in the browser offline: 2D floor plan, 3D build-up with furniture, natural light hour by hour and first-person walkthrough. All in a single HTML file.

  • Diffusion models for image generation

    Study and implementation of diffusion models applied to car synthesis with the Stanford Cars dataset. It includes tuning an existing architecture and computing FID to compare versions.

  • Foundations of statistics and classical learning

    Four assignments from the statistics part of my double degree: classification, segmentation of a real survey, housing price regression and multivariate analysis with principal components and factor analysis in R.

  • Agent harness for Pi and Orca

    A Pi extension that launches subagents as supervised terminals via Orca: declarative model routing, a hash-verified manifest and a diagnosis that found a duplicated language server tree.

See all projects

Recognition

Selections and competitions

Experiences that broaden my learning and connect me with other teams.

  • ProgrammeJuly 2026

    Innovation & AI Experience

    IE University and Santander Open Academy

    I was one of the 90 selected from more than 105,000 applications for two weeks in Madrid on innovation, entrepreneurship and applications of AI, together with 89 people from other countries and fields.

    My contribution: We learned to identify real needs, to question our own ideas before defending them, to build prototypes and to frame a business model around them.

    Santander Open Academy

  • Competition18-20 September 2026

    HackSpain 2026, Prosper AI track

    HackSpain, Madrid

    36 hours in Madrid to build a voice agent that answers a clinic's phone and makes real bookings. Team of five people.

    My contribution: I built the fake clinic for testing, the 21 scenarios across 18 failure families and the deterministic field-by-field comparator, so we could iterate without burning the challenge's scored minutes.

    The team's chronicle

  • Award2018 - 2019

    Andalusian Mathematical Olympiad

    Sociedad Andaluza de Educación Matemática Thales

    I qualified among the five best students of Almería for the Mathematical Olympiad of Andalusia. It was the first time I competed solving problems that were nothing like the ones at school.

    Scope and limitations
    • The qualification is provincial, not Andalusian.

    Olimpiada Matemática Thales

Track record

Experience

  1. February 2025 — presentnow

    Data Scientist, credit risk

    Grupo Cajamar · Almería

    Within the provisions parameters team under the IFRS 9 framework, I am responsible for calibrating, monitoring and projecting the LGD parameter (loss given default).

    • Extraction, transformation and cleansing of large volumes of data, with the outlier treatment required by the regulatory framework.
    • Analysis of recovery flows and calculation of the LGD parameter by portfolio segment.
    • Development and calibration of projection models under macroeconomic scenarios, with sensitivity analysis and performance tracking.
    • Assessment of the impact on provisions and preparation of the technical documentation in line with audit and governance standards.
    regulatedRSQLExcel
    More context

    Models and risk decisions

    I work on LGD calibration, monitoring and projection. The workflow combines historical data, recovery analysis, segmentation and macroeconomic scenarios.

    I have added automated quality checks, structured logging and reproducible documentation to support troubleshooting and review. I also prepare impact analyses and explain results to risk, business and validation teams.

  2. September 2025 — presentnow

    AI and Automation Engineer

    Self-employed · Remote

    I design and build bespoke AI automations and tools for internal company processes, almost always for people who have been doing the same thing by hand in a spreadsheet for years.

    • Development of automated flows within the Microsoft ecosystem: SharePoint, Power Automate and Azure services.
    • Digitising manual processes.
    AzureSharePointPower AutomatePythonLLMs
    More context

    Applied automation

    I design tools to extract information, summarise content and reduce manual work. Projects include OCR and LLM invoice extraction and a bot that turns site visit notes into draft reports.

    The workflows combine structured outputs, validation and human review before using the result.

  3. October 2024 — February 2025

    Data Scientist

    PwC, Deals · Granada

    Advanced analytics and pricing optimisation project for a Spanish transport operator.

    • Building the project database and the first pipelines, joining heterogeneous sources that did not speak to each other.
    • Development of demand forecasting models and dynamic pricing prototypes.
    • Optimisation of the purchases-per-visit ratio through predictive modelling of user behaviour.
    PythonpandasNumPyscikit-learnXGBoostPlotly
    More context

    Analytics for operations

    I developed pipelines and predictive models for a transport operator, alongside pricing and forecasting analyses. I prepared visualisations and results to support client discussions and operational decisions.

Toolbox

Knowledge in practice

From foundations to tools, and from tools to projects.

5 areas · 93 skillsA preview of the knowledge map: one dot per area, sized by how many skills it holds. It is only a preview; the full map, with level and freshness, has its own page.

Open the interactive map

A map of what I have studied and what I use. Each node shows its coursework and, where applicable, projects where I have applied it.

Explore all skills

Mathematics

Algebra, analysis, geometry, probability and modelling methods.

  • StatisticsInference, hypothesis testing, bootstrap, and interval estimation.Coursework: double degree
  • Linear algebraDecompositions, PCA, factor analysis, and the geometry of representation spaces.Coursework: double degree
  • OptimizationGradient descent, regularisation, and principled hyperparameter tuning.
  • ProbabilityProbabilistic models and stochastic processes applied to time series.Coursework: double degree
  • Differential geometryPseudo-Riemannian manifolds and the exponential map, the core of my thesis.Coursework: double degree
  • Abstract algebraGroups, rings, fields, extensions and Galois theory.Coursework: double degree
  • Computational algebraPolynomial ideals, Gröbner bases and symbolic computation.Coursework: double degree
  • Functional analysisNormed, Banach and Hilbert spaces, duality and operators.Coursework: double degree
  • Calculus and analysisLimits, series, differential and integral calculus in one and several variables.Coursework: double degree
  • Differential equationsOrdinary differential equations, systems and methods for studying their evolution.Coursework: double degree
  • Multivariate statisticsDependencies between variables, dimensionality reduction and multivariate analysis.Coursework: double degree
  • Statistical inferenceEstimation, confidence intervals and hypothesis testing.Coursework: double degree
  • Numerical methodsError and stability, systems of equations, interpolation and numerical integration.Coursework: double degree
  • Mathematical modellingDifference equations, population models, calculus of variations and reaction-diffusion.Coursework: double degree
  • TopologyTopological spaces, continuity, compactness, fundamental groups and coverings.Coursework: double degree
  • Complex analysisHolomorphic functions, complex integration and residues.Coursework: double degree

Computer science

Algorithms, computation, software, systems, networks and interaction.

  • PythonPrimary language for data, models, and services. Typed and tested.Coursework: master’s
  • GitHistory as documentation: atomic commits and branches with intent.
  • SQLAnalytical queries over large volumes and relational modelling with row-level policies.
  • TestingTests as a contract: property, edge cases, and metrics that fail when they should.
  • Software architectureDecisions recorded as ADRs, including the ones that abandon an idea.Coursework: double degree
  • RWorking language in credit risk: calibration, monitoring, and reproducible documentation.Coursework: double degree · master’s
  • Algorithms and data structuresComplexity, graphs, and data structures. Foundation of the double degree.Coursework: double degree
  • DockerReproducible packaging of services and evaluation environments.
  • Continuous integrationPipelines that lint, typecheck, test, and publish without intervention.
  • Linux and shellWorking environments, VPS services, and command-line automation.
  • TypeScriptFront-end and tooling, in strict mode.
  • Computer architectureProcessor organization, assembly, memory hierarchy and performance.Coursework: double degree
  • DatabasesRelational modelling, queries, data design and information systems.Coursework: double degree
  • Computability and complexityAutomata, languages, Turing machines and complexity classes.Coursework: double degree
  • Concurrency and distributed systemsSynchronization, process coordination and distributed computing.Coursework: double degree
  • Physics and electronics foundationsElectromagnetism, circuit theory and the devices underlying computers.Coursework: double degree
  • Computer graphicsGeometric transformations, scene representation and visualization.Coursework: double degree
  • Server engineeringConfiguration, performance evaluation and availability of web servers.Coursework: double degree
  • Human-computer interactionMobile interaction, gestures, motion and virtual reality environments.Coursework: double degree
  • Logic and discrete mathematicsLogic, proofs, combinatorics and discrete structures.Coursework: double degree
  • Object-oriented programmingAbstraction, encapsulation, dynamic memory and object-oriented design.Coursework: double degree
  • Computer networksNetwork architecture, protocols and communication between systems.Coursework: double degree
  • Operating systemsProcesses, memory, files, input/output and system services.Coursework: double degree

Data science

Applied statistics, machine learning and analysis across data types.

  • Machine learningFrom problem to metric: splits, validation, and leakage control.Coursework: double degree · master’s
  • Credit riskCalibration, monitoring, and projection of loss severity, under model governance.
  • Anomaly detectionRobust references, thresholds under a false-alert budget, and explainable suppression.Coursework: master’s
  • Model evaluationFrozen splits, a blind test, and metrics with their interval rather than one number.
  • IFRS 9The provisioning framework where the model has to pass an external audit.
  • NumPyVectorisation and numerical computing, including hand-rolled 3D geometry.
  • pandasData manipulation with explicit dtypes and traceable operations.Coursework: master’s
  • scikit-learnClassical modelling with cross-validation and hyperparameter search.Coursework: master’s
  • Deep learningArchitectures and training: diffusion, convolutional, and graph models.Coursework: double degree · master’s
  • Causal inferencePast only: a detector that never sees the future, so its alerts can be defended live.
  • LLMsUse under a strict contract: validated outputs and no direct write permission.
  • MLOpsTraining, versioning, and deployment with reproducible evidence for each run.
  • PyTorchPrimary framework for research and model prototypes.
  • Time seriesTemporal order, causality, and comparisons across laps or periods.Coursework: master’s
  • MLflowExperiment tracking: every published number points at a specific run.
  • Computer visionDetection, tracking, and segmentation over video and aerial imagery.Coursework: double degree · master’s
  • XGBoostGradient-boosted models for tabular data, with explicit preprocessing.
  • RAGRetrieval over first-party documentation and evaluation of answer quality.
  • Graph neural networksConvolutions over structures that are not a grid, including the non-Euclidean case.
  • YOLOReal-time detection, fine-tuned on a domain-specific image set.
  • Social network analysisCentrality, communities, diffusion and agent-based modelling.Coursework: master’s
  • Advanced machine learningImbalanced classes, multi-output, semi and self-supervised learning, transfer learning and AutoML.Coursework: master’s
  • Big Data and distributed processingLarge-scale storage, MapReduce, Spark, distributed analytics and federated learning.Coursework: master’s
  • Bioinformatics and omicsGenomic and protein data, alignment and computing for biomedical applications.Coursework: master’s
  • AI search and planningHeuristic search, constraint satisfaction, agents and planning.Coursework: double degree
  • Clustering and association rulesCentroid and hierarchical clustering, validation and association patterns.Coursework: master’s
  • Business intelligenceData warehouses, analytics and decision support systems.Coursework: double degree · master’s
  • MetaheuristicsTrajectory and population search, genetic algorithms and differential evolution.Coursework: double degree
  • Text mining and NLPText preprocessing, representations, opinions and sentiment analysis.Coursework: master’s
  • Data preprocessingMissing data, discretization, feature selection, regularization and PCA.Coursework: master’s
  • Knowledge transferCommunicating results, business models and developing data science applications.Coursework: master’s

Tooling

The concrete layer: platforms, libraries, and services I build with.

  • pytestTest suites with fixtures, property cases, and reproducible execution on pinned versions.
  • PostgreSQLTransactional database with versioned migrations and a tested restore path.
  • Next.jsWeb application with server rendering, actions, and scheduled jobs.
  • PlaywrightInterface tests in a real browser and automated screenshot capture.
  • SupabaseManaged Postgres with auth, private storage, and row-level access rules.
  • AstroStatic sites with islands: this portfolio, where only the three interactive layers load JavaScript.
  • FastF1Public F1 telemetry, gaps and format quirks included.
  • LiteLLMOpenAI-compatible gateway with virtual keys and per-project budget.
  • OpenCVReal-time video and image processing.
  • StreamlitAnalytical prototypes shipped in hours rather than weeks.
  • Three.js3D scenes in the browser, mesh export, and first-person walkthrough.
  • Apache SparkCoursework in distributed processing, Spark SQL and large-scale analytics.Coursework: master’s
  • Arize PhoenixLLM tracing and evaluation, self-hosted, over the open tracing protocol.
  • Hadoop and MapReduceThe Hadoop ecosystem and algorithm design under the MapReduce paradigm.Coursework: master’s
  • LiDAR and public dataDigital surface models and official cartography as the geometric source of truth.
  • MongoDB and NoSQLCoursework in NoSQL storage and MongoDB exercises.Coursework: master’s
  • Azure and Microsoft 365Automation in corporate environments: SharePoint, Power Automate, and internal flows.
  • BlenderHeadless command-line rendering inside a reproducible pipeline.
  • Google CloudData pipelines and analytics on managed services.
  • Power AutomateDigitising manual processes without writing a whole application.

Languages

Who I can talk to.

  • SpanishNative.
  • EnglishCambridge C2 certified. Documentation, interviews, and technical writing.
  • FrenchBasic.

Background

Education and certifications

Degrees

  1. September 2024 — August 2025

    Master's Degree in Data Science

    Universidad de Granada

    An intensive year of advanced statistics, machine learning, optimisation and distributed processing, capped with a master's thesis on remote sensing applied to agriculture.

    • Machine learning, probabilistic models, time series and big data.
    • Master’s thesis: olive grove monitoring with satellite, weather and field data.
    More context
  2. September 2019 — June 2024

    Double Degree in Computer Engineering and Mathematics

    Universidad de Granada

    Five years taking both degrees at the same time.

    • Algebra, analysis, geometry, probability and statistical inference.
    • Algorithms, systems, databases, machine learning and computer vision.
    More context

Certifications

  • Agents and LLMsHugging FaceDecember 2025
  • Machine Learning Engineer TrackDataCampDecember 2024 — January 2025
  • Data Engineering on Google CloudFundación AI Granada Research & InnovationSeptember 2023 — January 2024
  • Business Intelligence and analytics with LookerGoogle Cloud Skills BoostJanuary 2024 — June 2024

Languages

  • Certificate of Proficiency in English (C2)Cambridge English Level 3 CertificateJune 2019

A short break

One more lap

I also enjoy Formula 1. Try a qualifying lap here.

Qualifying lap

A two-dimensional F1 mini-game: trace the lap, brake in time, and let the three sectors decide.

Lap--:--.---
Best--:--.---
Delta—
Sector1/3
Speed0
Sector 1--.---Sector 2--.---Sector 3--.---

Qualifying lap

Throttle, brake and steer. Brake before the corner or you will lose grip.

No lap on record yet.

Keyboard: ↑ throttle · ↓ brake · ← → steer · space pause · R restart · Touch: the buttons below

Contact

Let's talk

If you have a data problem, a model that will not quite add up, or a role that fits, get in touch.

Terminal mode

german@portfolio: ~/portfolio26 cmds