Java Backend · React · Workflow Automation with LLMs on High-Traffic Platforms
↓ or just scroll down
My work spans the full engineering cycle: from designing Java/Spring services and database migrations to React components and automation scripting. I've built and maintained critical systems in multi-market environments with performance, availability, and stack-compatibility constraints that can't be ignored.
I work systematically: before proposing a change, I scope the exact boundaries of the problem, verify hypotheses with concrete evidence, and limit changes to the minimum required. I've incorporated LLMs as a regular work tool — not just for code generation, but for incident diagnosis, technical documentation management, and orchestrating experimentation pipelines.
I'm interested in the space where classical software engineering meets LLM systems: agent design, quantitative model evaluation, inference cost control, and observability. I'm building practical experience in MLOps and in integrating AI into real engineering workflows, beyond prototyping.
May 2017 — present
Idealista — national-scale real estate marketplace, multi-market European presence
August 2015 — May 2017
Darwinex — trading broker and platform
March 2015 — August 2015
HOMERIA Open Solutions
June 2013 — July 2015
Quercus Software Engineering and Research Group (University of Extremadura)
2011 — 2015
University of Extremadura
GPA: 8.23/10
How I fold LLMs into my actual workflow, with cost control and quantitative evidence.
Sustained design and maintenance, throughout my 9 years at the company, of the search and map-visualization subsystem for a high-volume real estate inventory catalog.
The challenge was combining Solr-based indexing and search relevance with efficient cartographic rendering over large volumes of geospatial data, keeping response times low at scale.
Automates duplicate detection and quality assessment in a high-volume image catalog, replacing manual review.
The challenge was designing a reproducible experimental protocol (fixed snapshot sets, A/B/C model comparison, unbiased metric calculation) with inference cost control.
A set of reusable skills, hooks, and agents that reduce the operational burden of recurring tasks: PR reviews, ticket management, documentation publishing, and periodic quality and SRE reports.
The challenge was designing clear component contracts and lifecycle hooks (pre-commit, pre-push) that run without intervention.
Recovers and structures documentation of business rules scattered across a long-lived monolith, transforming them into verified, source-linked Confluence pages.
The challenge was orchestrating code-review agents able to extract and verify business logic directly against the source code, through an iterative cross-validation process.
Prevents accidental exposure of exact property coordinates by substituting them with approximations according to business rules.
The challenge was designing a configurable coordinate-approximation layer driven by business rules, applied retroactively to already-migrated data without downtime.
Generates and publishes periodic SRE and quality reports (performance metrics + automated test results) to Confluence without manual intervention.
The challenge was chaining data from different monitoring systems and producing correctly formatted Confluence Storage Format content via a reusable skill.
A personal mobile app that locates the nearest public defibrillators (AEDs) and guides the user to them.
Personally maintained since 2015, focused on a minimal, fast experience for situations where every second counts.
Cáceres, Spain · 100% remote