AI Engineer · Generative & Agentic AI · Full-stack product development

I build AI systems and the applications around them: agents, APIs, user interfaces, data models and deployment workflows. Alongside my professional work, I develop personal products that turn practical problems and interests into working software.

LinkedIn · Public repositories

Some projects have public source code; others are developed in private repositories. The descriptions below explain what each project does, my contribution and who I built it with.

Built with Javier Lucía Marco, iFlySEO helps customers generate SEO content, manage their websites and schedule publication. The product combines a backend and background workers with a WordPress plugin, an embeddable browser SDK and a commercial account portal.

My contributions include the hosted panel and portal experience, administration tools, account and billing workflows, and fixes to scheduling and publishing behavior. I have also worked on an admin assistant with scoped, read-only account queries and on deployment isolation and reliability improvements.

This project brings together the parts of AI product development that interest me most: generation workflows, usable interfaces, account boundaries and visibility into usage and costs.

Stack: Python · FastAPI · Celery · PostgreSQL · Redis · PHP / WordPress · JavaScript · Next.js · TypeScript · Docker

Product website

A website builder I develop within the iFlySEO ecosystem. It turns a business brief into a structured, multi-page website that users can edit through blocks or natural-language instructions.

My work covers the generation flow, reusable templates, responsive previews, image uploads, editing and publishing. Draft revisions and published versions are separate, and the builder uses a shared renderer for editing and public pages. The service also includes account access, signed integrations, custom-domain verification and subscription-based hosting controls.

Stack: Next.js · React · TypeScript · PostgreSQL · LLM APIs · Docker

Built with Javier Lucía Marco, Arista connects people with professional mountain guides. The application includes activity discovery, maps, guide profiles, booking workflows and separate spaces for clients, guides and administrators.

My contributions include multilingual routes and interface copy, profile photos, messaging usability, application rebranding, database access hardening and dependency security fixes. I also work on the configuration and checks needed to maintain the application across environments.

Stack: Python · FastAPI · PostgreSQL / PostGIS · Next.js · TypeScript · MapLibre · Stripe Connect · Docker

A personal project combining cycling with software and AI. Climber runs in the browser and connects to a compatible smart trainer through Web Bluetooth and FTMS. As the rider progresses through a route, the app sends the corresponding gradient to the trainer and records speed, cadence and power.

I built the route editor and training interface, Bluetooth integration, activity history and deployment workflow. Routes can be imported from elevation-profile images using a vision model, kept private or shared with other users. The application also implements Strava OAuth and activity uploads, plus TCX export.

Stack: Python · FastAPI · JavaScript · Web Bluetooth / FTMS · OpenAI Vision · Supabase · PostgreSQL · Vercel

Source code · Web app

2025 · Approximately January–November · Built with Manuel San Román García.

Tutoría was an education startup project using generative AI and retrieval-augmented generation (RAG) to provide contextual tutoring based on specific learning materials. The platform connected with university learning systems such as Blackboard, bringing educational content into the tutoring experience.

I worked across the product and its technical infrastructure: a Next.js frontend, a Python / Flask API and Auth0 authentication, together with the RAG pipeline. Educational documents were processed and indexed using embeddings; semantic search retrieved relevant passages to give the LLM context for its answers.

I also built the AWS infrastructure around the application, with EC2 for the backend, S3 for documents and files, and RDS for the database. The result combined user accounts, document storage, retrieval and answer generation in a complete web application.

Stack: Next.js · Python · Flask · Auth0 · Generative AI · RAG · Embeddings / semantic search · AWS EC2 · S3 · RDS

GitHub organization · Source code maintained in private repositories.

A personal web application for running a World Cup prediction pool. Participants predict scores and the tournament winner; administrators manage entries, fixtures and results.

I built the participant and admin flows, scoring rules, rankings and automated result synchronization. Predictions close at kickoff and stay hidden from other participants until the match closes. Database access policies enforce user boundaries, while result corrections trigger score recalculation.

Stack: Next.js · React · TypeScript · Tailwind CSS · Supabase Auth · PostgreSQL · Vercel

Source code · Web app

A personal project that lets an AI agent act as the accountant for a Spanish freelancer's VAT returns: the quarterly model 303 and the annual model 390. I ask the agent for a return; it gathers the invoices with its own tools (folders, email), asks whatever it needs to know about my tax situation, explains the result and produces the official file ready to import on the Spanish Tax Agency's website. I review, sign and submit it myself.

The agent orchestrates, but every amount is computed deterministically by the MCP tools. They keep a local invoice ledger with validation and an audit trail, calculate each box of the return and generate the file directly from the Tax Agency's official record-layout spreadsheets, which they check for new versions before every return. When there is a doubt, the agent searches the official instructions for that year instead of answering from memory. Everything runs locally: the tool never files anything or handles certificates.

It works with Claude Code, Codex and OpenClaw as a plugin marketplace, as a standalone MCP server or from the command line.

Stack: Python · MCP (FastMCP) · SQLite · openpyxl · pypdf · Typer · Claude Code and Codex plugins

Source code

Arturito is my personal AI agent built on OpenClaw, with its own GitHub identity. I use it to explore practical agent workflows for research, coding, automation, documentation and maintaining context between tasks.

This is my own agent configuration and ongoing experimentation on top of OpenClaw. Its separate account makes the agent's identity and activity visible alongside my own work.

Arturito's GitHub profile

I work as an AI Engineer at VASS, developing generative and agentic AI solutions. My experience includes AI architecture projects for Bankinter and Abante Asesores, covering multi-tenant platforms, RAG pipelines, autonomous workflows and custom AI tools.

I hold a Bachelor in Computer Science & Artificial Intelligence from IE University.

• AI: LLM applications, agents, multi-agent systems, RAG, LangChain, LangGraph and MCP.
• Application development: Python, FastAPI, JavaScript, TypeScript, React and Next.js.
• Data and infrastructure: SQL, PostgreSQL, vector databases, Docker and CI/CD.
• Cloud experience: Azure, AWS and Google Cloud.

Outside of software, I enjoy skiing, cycling, teaching and traveling. Several of these projects started with a problem I wanted to solve myself.

Feel free to connect with me on LinkedIn to discuss AI engineering, product development or any of these projects.