Senior Backend & AI Systems Engineer
Java · Quarkus · GraalVM native · Agent infrastructure (MCP, RAG, harnesses) · Next.js
-10B981?style=flat-square){kind=icon}
▓▓▒▒░░ · · · · · · · · ░░▒▒▓▓
I'm a senior engineer from Murcia, Spain, working as an independent contractor. Most of my career has been spent inside Java distributed systems — Capgemini, Accenture, and then TuLotero — and the last couple of years putting agentic AI into production instead of into demos.
I like the unglamorous half of AI engineering: invariants, guardrails, harnesses, regression tests with mutability, and the boring plumbing that decides whether an agent is a product or a party trick. If something I build ends up being useful to someone else, even better — that's why most of it ends up here in the open.
Right now I'm CTO & Tech Lead at Javadaba and Legally, and I build in public whenever the work allows it.
░▒▓ ─────────────────────────────────────── ▓▒░
| Project | What it is |
|---|---|
| fluxor-ide | An open-source agentic IDE built on step-DAG flows — roles, mods, triggers, MCP client integration, backlog cards and a notification center. Flows can be containerized and deployed as detached apps exposing MCP over Streamable HTTP, so agents can talk to each other. |
| cognitive-skills | A framework that extracts a person's cognitive profile and generates a personalized set of skills from it. |
Also cooking: RegWatch, a regulatory radar for Spanish normative sources (BOE, AEAT, TGSS, SEPE) on Drools/Kogito · a dithered component library where nothing is an SVG curve and everything is composed from atoms · a pay-per-use API catalog distributed over MCP.
░▒▓ ─────────────────────────────────────── ▓▒░
Selected production work
- TuLotero (15M+ users, high-concurrency distributed system, #1 downloaded app in the Spanish App Store during the Christmas season) — built the public MCP server; ran the pSEO + GEO + MCP initiative directly with the CEO; integrated a new payment provider from scratch with event-based rollback; built the SPEI microservice on Quarkus (~40k transactions/day); distributed + local Redis caching against a WildFly monolith. Proposed the AI workflow adoption that started internal AI use at the company.
- Grupo DIA (Accenture) — 15+ Quarkus microservices in a DDD domain behind a Spring API gateway with caching and multiplexing. Built the payment-methods and customer-addresses services from zero. Saga pattern, Liquibase, Pub/Sub, SQS, SES, BigQuery, testing with Mockito and Karate.
- BMW Group / Alphabet (Capgemini) — vehicle shipment management system; first international Scrum team, ended up coordinating it to hit the deadline.
- Santalucía (Capgemini) — legacy C++/Java microservices against IBM Informix, JSP backoffice, heavy reporting SQL, high L4 ticket volume.
Numbers I can point a commit hash at
- +58% throughput (330 → 520 rps) and p95 295 ms → 162 ms on a production web app
- GraalVM native unlocked: 0.669 s startup, −51% build time
- −98% serverless invocations and −75% compute on an audio fingerprinting pipeline
- ~68,000 programmatic SEO routes shipped with 0–10 ms TBT and CLS 0
- God component refactored 12,077 → 1,802 lines; E2E suite 4m12s → 48s
- Unauthenticated
/mcproute found and closed, documented with a failing test first
These come from an empirical pass over my own git history — traceable to commits, and separated from anything estimated. Plenty of that code was co-authored with coding agents; I don't count it as hand-typed production.
░▒▓ ─────────────────────────────────────── ▓▒░
Languages
Backend & runtime
Spring Batch · EJB · REST · microservices · Domain-Driven Design · API gateway · saga pattern · event-driven architecture · pub/sub
AI & agents
LLM integration · controlled chain of thought · regression testing with mutation · AI-assisted delivery workflows · clustering algorithms
Frontend
Angular / AngularJS · Java JSP
Data & messaging
IBM Informix · Google Pub/Sub · Amazon SQS / SES · Firebase Cloud Messaging
Cloud & DevOps
Lambda · S3 · SQS · SES · Cognito · CDN · App Engine · Cloud Functions · Pub/Sub · Storage · OAuth · CI/CD pipeline design
Quality & observability
Mockito · Karate · quality gates · code review · L3/L4 support · on-call incident response
Payments & integrations
Banking file processing (fixed-position mappers) · programmatic SEO
░▒▓ ─────────────────────────────────────── ▓▒░
Harness engineering as an emerging discipline — the harness as the deterministic partition of the system, while context behaves as a writable partition prone to rot. Poisoned code: context poisoning, codebase poisoning, adversarial poisoning, and what a real detection posthook looks like. Neuro-symbolic invariant inference. And, further out, what an AI-native OS and its UX should actually feel like.
