I''m preparing amazing content about development, technology and programming. Come back soon!

Thinking out loud about software, AI and development

Blog

Corporate website for Kumux built with Gatsby and JavaScript.

B2B e-commerce for an industrial safety products distributor, built with PHP and PrestaShop.

Internal app for lighting designers, built in pure React.

E-commerce for a subscription box brand targeting camping and outdoor enthusiasts.

Experience

Building plugin workflows and platform SDKs for a global travel commerce company. Working on developer tooling, API integrations and frontend performance for products used by airlines, hotels and travel agencies worldwide.

Telemedicine platform with AI-assisted diagnostics built for the Chilean health system. Developed with React, Gatsby and Strapi.

Previous personal portfolio and services website built with Next.js and TypeScript.

SaaS platform for restaurants: QR menu, table ordering and integrated payment. A freelance project that taught me a lot about product and clients.

Currently at Travelport. Previously shipped MVPs for startups from scratch.

Software engineer building weird systems, AI agents and fast interfaces.

Tutorials and experiences

Let's talk about projects

Let's Connect

Latest post

Technologies I use day by day

Hello, I am a Full-Stack developer specialised in Frontend development with high knowledge in Digital Marketing.

I am currently a full stack in TravelPort. I like to create new products and help people to develop their ideas. And when I am not online, I like to play sports.

The page you're looking for was not found.

I was born in Barcelona at the end of the year.

🐣 Birth

I got my first video console as a gift. A Sega Mega Drive!

🎮 First video console

I got my first Gameboy as a gift and with it I became increasingly passionate about technology.

🎮 Gameboy

I have my first computer: A Pentium 4 with 256MB of RAM + 📞 Internet. But only from 6 PM onwards and leaving my parents without a phone.

⌨️ First computer

DSL arrives and with it WoW, Counter Strike and LAG :(

🌐 DSL Internet

🏫 ESO Completed

I finish ESI, "Computer Systems Operation".

🏫 ESI Completed

I work at Zimbra in Italy as an Erasmus student. I start working at Punt Informatic.

🇮🇹 Erasmus & First job

I start at Grafix Gestió Informàtica and take Android and Java courses.

💼 Grafix & 📚 Courses

I finish the CFGS for DAI (Development of Computer Applications) at IES Carles Vallbona, Granollers.

🎓 DAI Completed

I start being a "Runner" and two great colleagues arrive at the company: <a href="https://www.linkedin.com/in/mart%C3%AD-duran-69257374/" target="_blank" rel="noopener noreferrer" style="color: #4A90E2; text-decoration: underline;">Marti Duran</a> and <a href="https://www.linkedin.com/in/marcgirbau/" target="_blank" rel="noopener noreferrer" style="color: #4A90E2; text-decoration: underline;">Marc Girbau</a>.

🏃 Running + Colleagues

🐶 La Menta

🏃 First 15k race

I start with CrossFit and investing in cryptocurrencies ₿.

🏋️ CrossFit & Crypto

I do a mountain half marathon and a 10k. I travel to Albania 🇦🇱 where my perspective on how the world works changes.

🏃 Half marathon & ✈️ Albania

First competition and injury in crossfit. Creation of the Gametravel brand in the university incubator in Mataró.

🏋️ CrossFit & 🚀 Gametravel

I finish a UOC course taught by <a href="https://www.linkedin.com/in/oriolrius/" target="_blank" rel="noopener noreferrer" style="color: #4A90E2; text-decoration: underline;">Oriol Rius</a> on IOT. I join Kumux as Frontend Developer — building React visualizations for colorimetric data and launching the company website with Gatsby.js. COVID-19 keeps us in quarantine. 🏠 #StayHome.

🎓 IOT & 🚀 Kumux & 🦠 COVID

I leave Kumux and go freelance under the freak.wtf brand. I lead end-to-end MVP development for startups and build Pideme.es — a QR-based ordering system so restaurant customers can order and pay from the table. Stack: React, Next.js, TailwindCSS, Node.js, Strapi.

💼 Freelance & 🚀 Pideme.es

I join Travelport as Software Development Engineer. I start on the SDK & Plugin Framework team building developer SDKs for booking APIs (flights, trains, hotels) and CI pipelines for partner plugins. I meet great people and grow a lot.

✈️ Travelport & 👥 New friends

I finally fulfilled my dream of seeing the northern lights. An incredible trip to Iceland that will be forever etched in my memory. ✨ Aurora Borealis magic!

🏔️ Iceland vacation

We move closer to the city. An important change that brings us closer to new opportunities and urban experiences.

🏠 Closer to the city

Pam arrives, a mini toy poodle that has us in love. I finally buy a new computer designed to work with AI - I had been using my old computer for almost 20 years, it was really ready to retire.

🐩 Pam arrives & 💻 New AI computer

A journey through the most important moments of my life, from my first steps in technology to where I am today.

My Timeline

Landing to help raise money for children in hospital

SaaS platform for restaurants: QR menu, table ordering and integrated payment.

B2B e-commerce for industrial safety products distribution

Internal web platform for lighting designers at Kumux

E-commerce store built to learn Next.js, deployed on Vercel

React Native app listing live cryptocurrency prices from a public API

E-commerce for a subscription box brand targeting camping and outdoor enthusiasts

Telemedicine platform with AI-assisted diagnostics

Portfolio and professional services website as a developer

JavaScript + Selenium script to auto-book CrossFit classes when a slot opens

Weather app built with React, Redux and Jest to practice modern state management

Learning projects

B2B e-commerce for an industrial safety products distributor. Built with PHP and PrestaShop, with custom catalogue management, bulk pricing and a client portal for recurring orders.

Internal web platform for lighting designers at Kumux, built in pure React. Allows teams to create, configure and visualise custom lighting products in real time before production.

Full e-commerce store built as a hands-on project to learn Next.js. Includes product listings, cart management, and checkout flow. Deployed on Vercel.

Mobile app built with React Native that lists live cryptocurrency prices and market data, pulling from a public API. A practical project to explore cross-platform mobile development.

Telemedicine platform built for the Chilean health system, leveraging AI to assist doctors in early diagnostics. Developed with React, Gatsby and Strapi as a headless CMS.

Corporate website for Kumux, a lighting innovation company based in Barcelona. Developed with Gatsby and JavaScript, focused on performance and visual storytelling to communicate the brand values.

E-commerce platform for a subscription box brand targeting camping and outdoor enthusiasts. Built with PHP and PrestaShop, with custom subscription logic and seasonal catalogue management.

Telemedicine platform with AI-assisted diagnostics. Doctors can review patient data, run AI-powered analysis and issue remote consultations. Built with React, Gatsby and a Strapi backend.

Personal portfolio and services website built with Next.js and TypeScript. Showcases my experience, projects, and technical skills. Optimised for performance, accessibility and SEO.

SaaS platform that lets restaurant customers scan a QR code, browse the menu, place orders and pay directly from their table — no app download required. Built with PHP and Laravel, integrated with kitchen display systems.

Automation script written in JavaScript using Selenium to monitor available slots in my CrossFit gym and automatically book a spot when one opens up — so I never miss a class.

Software Development Engineer at Travelport since 2022. Building plugin workflows for a global travel commerce platform operating in 165+ countries, with a focus on the frontend of the Search form.

Weather forecast app built to practise React with Redux for state management, Jest for unit testing, and the useEffect hook for async API calls to a public weather API.

Fundraising landing page created to support an event raising money for hospitalised children. Built with HTML, CSS and Tailwind CSS, focused on emotional impact and conversion.

Fundraising landing for hospitalised children


A few months ago my team shipped a change that worked perfectly. Everything looked fine until another team shipped something on their end and a bug appeared — nothing obvious, the kind that takes time to surface and doesn't point clearly to its origin.

When I started investigating, I delegated to the AI. It gave me four solutions. All coherent. All technically valid. None of them quite fit.

We got together with the other team for a pair programming session. In twenty minutes we found the real solution — one the AI never proposed, because it required understanding a design decision that neither team had documented separately.

That experience left me thinking about something I've been seeing more clearly for months: AI multiplies productivity, but it can also make you lose control over what you're building. I don't think the problem is AI. I think the problem is how and when we choose to delegate to it.

> The AI always gravitates toward the most obvious option. Technically valid, but not always correct for your context.

The gap between what we expect and what actually happens:

![Expectations vs Reality: AI will understand the full context of the system vs AI doesn't know what was decided in a meeting back in 2019](/images/contents/ai-expectations-vs-reality.png)

## When AI works incredibly well

There are tasks where the improvement is absurd: boilerplate, refactors, tests, documentation, initial debugging, quickly exploring solutions.

**AI amplifies when...**
- You understand the system well
- The problem is documented
- You can evaluate the output

**AI fails when...**
- Context is cross-team
- The decision isn't written down
- "Valid" feels like enough

## The real risk isn't AI failing

The real risk is more uncomfortable: trusting solutions we don't fully understand. AI produces responses that are coherent and technically valid, and now you can produce enormous changes in minutes. But you modify things fast without truly understanding them, and that generates a false sense of comprehension.

> "Valid" doesn't always mean "correct for your context." And the less you understand the system, the more you depend on AI's judgment to know the difference.

Me, merging PRs at 11pm:

![This is fine: me accepting AI-generated changes while production burns](/images/contents/ai-this-is-fine.png)

## What AI doesn't see

AI can read code, but it doesn't understand why certain decisions exist, what technical debt can't be touched, or what the team next door knows that was never documented.

That knowledge lives in people, conversations, and past incidents. In complex systems, the best decision isn't usually the most elegant one — it's the one that fits everything that isn't written down.

## The new value isn't writing faster

> It's understanding better.

AI automates, accelerates, and proposes. But judgment is human: deciding what to build, what not to touch, what risks to take on.

When context is fuzzy, you start accepting answers too quickly. You delegate decisions without realizing it.

The cost of producing software drops. The cost of understanding the problem doesn't. And that context still lives primarily in people. Sometimes, at the intersection of two teams in a twenty-minute meeting.

---

*By the way: this post was written with AI. The irony isn't lost on me.*


AI and productivity: the hidden cost of delegating control


# Building a Safety Net Around DeFi Automation

For the last months I’ve been building a safety layer around DeFi automation.

It doesn’t farm, it doesn’t chase yields, it doesn’t try to be “smart money”. Its only job is much more boring: **watch things, connect a few dots, and say “this is safe enough to act” or “please stop”**.

This post is a small introduction to what I’m building and, more importantly, how I’m thinking about it. I’m *not* going to show you everything – on purpose.

## What problem am I trying to solve?

If you’ve been around DeFi automation for a while, you’ve probably seen at least one of these patterns:

- Scripts that keep running happily long after the original assumptions are dead.
- “Health checks” that only verify that a process is up, not that it’s safe.
- Bots that will try to rebalance or harvest in environments that are clearly hostile: no gas, bad pricing, broken RPCs.

What I want from this safety layer is simple:

> Before you touch funds on-chain, answer one question:  
> **“Is it actually responsible to act right now?”**

Sometimes the answer should be “yes, go ahead”.  
Sometimes the answer should be “no, and here’s why”.  
And sometimes the answer should be “I have no idea, so I’m going to block you”.

That last case is the one I care about the most.

## A tiny bit about how it works (without giving away everything)

Very high level:

- It watches different kinds of signals:  
  balances, gas availability, strategy health, exit conditions…
- It combines them into **explicit decisions**:
  - “act now”,
  - “wait and re-check later”,
  - “refuse to act under current conditions”.
- It tries to make the **reasons** visible:
  - not just “failed”, but “skipped because insufficient gas runway”,
  - not just “error”, but “RPC is blind, treating this as unsafe”.

Under the hood it’s just TypeScript, a bunch of services, and more tests than I’d like to admit. The interesting part is not the framework, it’s the questions I force the code to answer.

There are parts I’m not going to describe in detail:

- exact thresholds and combinations of signals,
- operational internals,
- some of the production wiring.

Think of this series as **field notes**, not a full spec.

## What this blog series will focus on

I don’t want to turn this into a changelog of features. I’d rather write about the decisions behind them, for example:

- When should an automated system **refuse** to act, even if it technically can?
- How do you decide what “enough runway” means before you’re being irresponsible?
- Where do you consciously trade developer comfort for user safety?
- How do you encode those ideas in code so future-you can’t silently ignore them?

Each post will probably zoom into a small piece of the system:

- a service,
- a pattern,
- a specific check or threshold,

and treat it as an excuse to talk about **good vs bad code, and good vs bad decisions**.

You’ll see real snippets from the codebase, but always with a bit of blur where I don’t want to go into operational details.

## What I’m willing to show (and what I’m not)

I *am* willing to show:

- how I design health checks that do more than “is the process up?”,
- how I think about failing *closed* vs failing *open*,
- how I keep myself honest with tests and explicit contracts.

I’m *not* going to show:

- everything that can be configured and how,
- every single signal and threshold,
- anything that would meaningfully leak operational posture.

If sometimes a piece of code looks a bit redacted or simplified, assume it’s deliberate.

## What’s coming next

In the next post I’ll zoom into a piece that sounds trivial and really isn’t:

> how I decide whether there’s “enough gas” for the system to act at all.

On paper it looks like “just check the balance”.  
In practice it becomes a conversation about **runway, RPC failures, and when being blind should mean a hard stop**.
