About

I turn data, models, and software into systems people can use.

I’m Albert Queraltó, a PhD-trained data scientist and software developer based in Barcelona. I work across machine learning, data engineering, backend systems, and product development, with a particular interest in systems that have to remain useful after the notebook experiment is over.

My work combines scientific reasoning with software engineering. I care about how a model is validated, but also about how data reaches it, how predictions are exposed, how failures are observed, and how a user ultimately makes a better decision with the result.

That approach is visible in projects such as Payrithm and TenderWise, where machine learning, deterministic rules, APIs, background jobs, databases, and user-facing workflows have to work together as one product.

This portfolio documents the architecture, trade-offs, mistakes, and operational lessons behind those systems. The goal is not to present isolated demos, but to show how I reason about production data and software problems end to end.

Areas I work across

Machine-learning systems

Time-series modelling, anomaly detection, supervised learning, validation, explainability, model serving, and production monitoring.

Data engineering

API and file ingestion, ETL workflows, PostgreSQL and PostGIS, data quality, scheduled processing, and operational data pipelines.

Backend and SaaS

FastAPI services, multi-tenant application design, Celery workers, Redis, permissions, auditability, and decision-oriented product workflows.

Production engineering

Docker, Linux, Nginx, resource-constrained deployments, observability, reliability, and pragmatic infrastructure choices.

How I approach technical work

  • → Start from the decision or operational problem, not from the model.
  • → Validate what was actually knowable at prediction time and make leakage difficult by design.
  • → Prefer observable, maintainable systems over unnecessarily complex infrastructure.
  • → Separate domain rules from probabilistic models when the distinction matters to users.
  • → Document failure modes and trade-offs, not only the final architecture.

Background

  • → PhD-trained problem solving and scientific methodology.
  • → Academic background connected to Universitat Autònoma de Barcelona and Universitat Oberta de Catalunya.
  • → Based in Barcelona, Spain, and open to data science, machine-learning engineering, software engineering, and product-oriented technical work.

Work with me

If you are building a data-heavy product, an ML-backed decision system, or a production workflow that needs to become more reliable, I’m interested in the engineering decisions behind it.