Agentic AI that reaches production — and the client.
I'm Sergi, a software architect based in Barcelona. I work at the intersection of infrastructure, delivery and developer experience, designing the platforms and tooling that let teams ship reliably, repeatably and efficiently.
My bias is toward results in production and the client's real context, not architecture for its own sake. Short feedback loops, infrastructure as code, and pipelines that do the boring work so people don't have to.
Currently a full-time Software Architect at GPAINNOVA, Barcelona — and I also take on independent projects on the side.
sergi.fígols — barcelonaSeven areas where I bring the most impact — from the metal up to the product.
Support designing and shipping agentic AI — LLM-powered agents, tooling and orchestration wired into real products.
Reproducible, version-controlled infrastructure — provisioned as code, never by hand.
Self-hosted GitLab on AWS with automated build, test, scan and deploy on every commit.
Experiment tracking, model lifecycle and heavy training on interruptible Spot compute.
Scalable, secure deployments on AWS — EC2, VPN access and containerized services.
Production web apps end to end — Next.js front ends over robust Python back ends.
Connecting machines and devices — communication systems and telemetry for Industry 4.0 and 5.0.
Two platforms I built, and the products they power. Pick a case from the index.
Teams shipped by hand — builds were inconsistent, releases were slow, and there was no shared delivery standard across repositories.
A self-hosted GitLab on AWS with reusable pipeline templates: build, test, scan and deploy on every commit, promoted through environments with a GitOps workflow.
Model training was ad-hoc and expensive — experiments weren't tracked, results weren't reproducible, and GPU time ran on costly on-demand machines.
An MLOps platform around ClearML for experiment tracking and orchestration, running heavy training on AWS Spot instances, with a model registry feeding the serving layer.
APS runs automated luggage-cart rental systems in airports worldwide — Costa Rica, Turkey, Milan. Travelers rent carts with local currency and are refunded on return.
A real-time MQTT bridge between their logical control devices and the payment processing systems — plus cloud infrastructure in Django and frontend interfaces for the local machines.
A basketball club needed a public web presence and a way for non-technical staff to manage their own content.
A public website and backoffice — CMS-driven pages with an API powering both the public site and the admin tooling the club uses day to day.
Restaurants get customer reviews, but there's no easy way to credit the waiter who actually earned each one.
An app where every waiter carries an NFC card; customers tap it to leave a review tied to that waiter, who then earns benefits, prizes and recognition.
ICT Systems Engineer, UPC — Universitat Politècnica de Catalunya.
After GPAINNOVA internalized me in February, I moved into the architecture role — defining and shipping to production the infrastructure, CI/CD and the MLOps platform.
Embedded with GPAINNOVA through Between Technology — working across diverse projects, from IoT over MQTT to backend services with Django and Odoo, and contributing to the GitLab CI/CD pipelines.
Full-stack developer on diverse projects — from frontend with React and Next.js to backend services in Python, AsyncIO and Kubernetes.
While studying at UPC — resolving computer and device issues, and later managing the VMware infrastructure.
Notes on engineering, architecture and how things get built. More coming soon.
A walkthrough of the stack behind this portfolio — Next.js, Tailwind and next-intl — and where I'm taking it next.
Read article →How to build an agent for Odoo that talks to the different modules — currently living in one of my repositories.
Coming soonRendering very heavy 3D objects on the server so they show up on small, low-powered screens.
Coming soonHave a platform to stand up, a pipeline to tame, or models to get into production? Let's talk.