Context

For almost 6 years I was involved in ideation, development and handoff, covering the design needs of the eCommerce Logistics and Finance verticals.

Outcome

It would be hard to explain everything that happened over those 6 years, but it can be summed up as more than 11 projects in the eCommerce logistics backoffice, among them:

• Route maintenance and dynamic courier selection
• Setup and maintenance of stock sources
• Order returns operations at returns centres
• Unified dashboards
• Management of product injections into markets
• Real-estate management system for the company
• Contract management for models, props and photographers
• Maintenance management for buildings and distribution centres

What stands out?

I've been the go-between for the needs of the business, users, POs and the development teams. Understanding the problem and empathising with the client and the users was both a discovery and a lesson while carrying out the projects. Having been a frontend developer let me speak the same language as the development teams. I also worked with the other designers in the UX team to adopt the team's methodology, and on rolling out and contributing to the design system.

Challenge

B100, a neobank, only had a native app for onboarding customers to its online bank. It needed a responsive web flow alongside it, for both desktop and mobile.

What I did

We didn't know how B100's sign-up worked internally; we did know its brand, its look and the mobile flow. I started by looking at how other banks do it (Revolut, Abanca…) and mapped out how they should work at B100. That gave five validation steps, plus other variations coming from special cases for Abanca customers (B100's parent bank).

I documented the client's visual style and built the basic form components before designing any screen.

How I used AI

Before opening Figma I asked ChatGPT and Gemini for a sequence diagram of the sign-up. Development used it to see the whole interaction with nothing left open to interpretation; I used it as the basis for the prototype I was going to build to validate with the client.

With the mockups, the prompt and the diagram I generated a clickable prototype of the whole flow in Figma Make. Once validated, I went back to Figma to turn it into a document detailing the remaining flows and corner cases.

What didn't work

It works for a simple flow; for complex or very open-ended ones it doesn't. I had to keep the prototype to the happy paths and leave the error cases for the Figma work. Every generation gave a different proposal, and moving the prototype into Figma was an extra step (these days that's no longer an issue).

Outcome

The first MVP, from briefing to validation, was ready in one week of work. The site was live for around four or five months although, for reasons I don't know, it was taken down afterwards. I never got usage data, but I would have measured the drop-off rate at each step, especially the liveness check and the ID upload, and how many sign-ups came through the web versus the app.

Challenge

Deportivo de A Coruña needed to speed up content maintenance for the native iOS and Android apps while the club rolled out a global CMS for the organization.

Solution

A tool that lets the club's content managers make changes easily.

Beyond managing the squad and teams, I proposed scheduled management of the apps' advertising banners.

What I did and how

I designed the information architecture around the club's needs:

• Team and player maintenance
• Banner management
• Banner scheduling
• Promoted product management

From there I made a preliminary design and a clickable prototype, and built the proposal with Claude Code.

Outcome

A responsive web tool, built in very little time, to meet a specific need of the club.

Challenge

Design a product for managing push-notification campaigns.

What I did and how

I held working sessions with the stakeholders and the people who knew the domain and context of the future tool. I turned the main flows and use cases into proofs of concept and validated them with those people.

Starting from an in-house design system, I built the components needed to show the notification schedule, with these decisions:

• Three breakpoints: mobile, tablet and desktop
• Components designed mobile-first
• Reusing them as much as possible to lighten the load on development

Outcome

Alongside the Figma document detailing the use cases, I built a project library with the components it needed, and some were promoted into the company's design system.

We kept talking with the development team throughout, which made development easier by reusing components across screen sizes.

Challenge

Build a native iOS app, from Figma designs and with AI, to show what Céfiro can do. It's a very basic app, meant as sales support when explaining the product.

Solution

It was an experiment to validate the workflow from a Figma design to a working tool, or at least a very high-fidelity prototype, using AI.

What I did and how

I defined two use cases: a notification inbox and a map to validate geofenced notifications.

Visually, the app had to follow Apple's HIG, specifically the Liquid Glass style. I designed it in light and dark mode and built the app in SwiftUI with Claude Code.

Outcome

A working native iOS app, ready for development to finish connecting it to the relevant services.

Challenge

PAS GARAGE sells and sources distinctive vehicles across Spain and Europe: unusual configurations, special editions and limited-production units. It needed a website that conveyed that judgement and presented each car as a collector's piece.

What I did

I designed the site in Figma, for desktop and mobile: a home page, a catalogue of units and a page for each vehicle with a photo gallery and viewer.

I built it as a static website, with no dependencies or build step. The Figma desktop grid is translated into fluid measurements, so the design scales between sizes without jumps. It includes a loading screen with a background animation on desktop, a form that sends straight to the client's inbox, and a cookie notice that loads no analytics until it is accepted.

Outcome

A four-page website now live, fast and easy to maintain: updating it is just a matter of copying the files to the hosting.

Challenge

A company from A Coruña with almost 20 years in forestry management (timber sales, land clearing, planting, and buying plots and inheritances) wanted to win customers online without signing up for a Google Business Profile, by its own choice.

What I did

I designed and built a home page and one page per service, each aimed at a specific local search, such as "land clearing in A Coruña".

Every page keeps the quote form in view, with direct access to WhatsApp, structured data for Google, a sitemap and a cookie policy that doesn't turn on Google Analytics until it is accepted.

What didn't work

A later SEO audit found that the service pages were too short, at 300 to 345 words, to compete without a Google Business Profile, and that the home page video was too heavy on mobile. The next step is to expand them with FAQs, service areas and regulations.

Outcome

A five-page website now live, built to rank each service separately in local searches.

Challenge

On every website I deliver, I check the SEO by hand, page by page: titles, descriptions, headings, alt text, canonical, sitemap. The usual tools return long lists of warnings without showing where each problem sits on the page, which makes the review slow and hard to show to the client.

What I did

I designed and built a tool that takes a URL, crawls up to 30 pages of the same site and shows each one as a real thumbnail. Every heading and image appears as a numbered pin on the thumbnail, linked to an editable card in a side panel.

It measures the HTML as a search engine that doesn't run JavaScript would read it, and it respects robots.txt. For each page it gives a score and a technical diagnosis: title and description length, canonical, language, favicon, heading hierarchy, alt text, Open Graph and structured data. It is read-only: proposals are exported as JSON or as snippets ready to paste into the <head>.

Outcome

I use it to audit the websites I design, before and after they go live. I used it to review the SEO of this portfolio, which was missing a sitemap, canonical tags, Open Graph and structured data.

Challenge

The Noroeste Estrella Galicia Festival runs for five days, with stages spread across A Coruña and overlapping gigs. The official calendar is one timetable per venue: you can't see what clashes with what, and it doesn't tell you how to get from one stage to another.

What I did

The core piece of this webapp is a horizontal view of the lineup by day, venue and time, where clashes are visible at a glance. Every gig can be tapped: it opens the band's profile, with Spotify data, and the walking route to the stage.

Around it I added favourites, sign-in with a password or Google, and "My route", a personal itinerary you can share and comment on. Since it would have real users, it includes GDPR consent with data export, a backoffice and Google AdSense ads.

How I used AI

I used Claude Code to generate both the horizontal timetable view and the routes between stages, and across the rest of the stack: Node.js/Express and SQLite on the backend, framework-free JavaScript on the frontend, deployed on Railway.

What didn't work

The routes. The first version didn't work out the walking path: it drew a straight line from point A to point B. I fixed it with OpenRouteService and kept refining until the routes were right. In the end I precalculated and cached them.

Outcome

The horizontal view worked: nobody needed it explained.

173 people used it and around 20 signed up, with no promotion at all and the app published only a few days before the festival. The response was very positive. It was all built in six days: 58 commits between July 29 and August 3, 2026.

The project carries on for next year's edition.

Challenge

Poker dice (A, K, Q, J, 8, 7) is a very traditional game in Galicia, and there was no app or game that recreated it. I wanted to bring it to mobile so people could play in real time with others, with dice that behave like real dice.

What I did

I built an app and a webapp for multiplayer matches, or a simulation against bots.

Around the game, its own economy: a virtual currency (Bules), a shop for skins and cosmetic items, and tournaments with restricted access. You can sign in with Google or play as a guest.

How I used AI

I used Claude Code to choose the game engine and the physics engine. Unity and Godot were too heavy for what I wanted, and the other engines either lacked the quality I needed or their physics didn't work well. I went with Three.js, which works very well on the web and fit what the game needed, and Rapier3D for physics.

I built the rest with Claude Code too: Node.js/Express, React and Socket.io for real time, the Android app packaged with Capacitor, push notifications through Firebase and a React backoffice.

What didn't work

The physics of the dice and the rolls: it had to be retuned to avoid clipping bugs, such as dice going through the table or into each other. The calculation system also had to be redesigned and centralised on the server, so the roll is sent out to every player and shown with the same physics and animation.

Outcome

It's already published on the Play Store and runs on the web, as a PWA and as an Android app, with backend, backoffice and monetization ready. A group of 30 players is now testing it to polish the mechanics and fix roll errors.

Challenge

GRISES is my black-and-white architectural photography project, signed as Ollofoto: a journey through series of buildings in different cities. Playing with perspective, vanishing points, symmetry and negative space, it sets out to change the way we see them and turn them into impossible geometries.

What I did

I took the photographs and designed and built the website to show them: a home page with the series, a gallery that can be filtered by city (Madrid and Coruña), the project page, a contact form and an English version.

Outcome

Two series published on the website, Madrid and Coruña.