For my physics degree thesis, I developed an easy-to-use simulator for Bose-Einstein Condensates utilizing a GPU-accelerated system with JAX (by Google). As part of the project, I also designed and implemented a web API to enable running simulations on the cloud, making the tool more accessible and efficient for researchers and students.
Physicist · turned · AI Engineer
Hi, I'm Marc Parcerisa
I'man AI Engineer
A physicist turned AI engineer and software developer. I've worked on everything from quantum-mechanics simulations to computer vision and natural language processing. These days I'm an AI engineer at Kabilio, building AI-powered accounting software, and finishing my studies in Data Science. Always chasing the next thing to learn.
Things I've built
Projects
I sometimes play DnD with friends, and wanted to create an interactive world map that I could project onto the table. To achieve this, I developed a web app that allows me to create and edit maps, add creatures, and control both their movements and the visibility of different map elements in real time. A backend server manages the game state and map data, while the frontend, built with NextJS, enables seamless real-time updates for an immersive gameplay experience.
A project I built to get hands-on with Retrieval-Augmented Generation (RAG) — the architecture that grounds a language model's output in relevant information pulled from a knowledge base, rather than relying on the model's parameters alone. I used it to put together a simple book recommendation system that retrieves candidate books and lets the model reason over them.
I'm now finishing a MSc in Data Science — currently working on my thesis. The degree covers a wide range of topics, from machine learning and deep learning to cloud computing and big data, with a strong focus on practical applications and projects, so be sure to check out my compiled list of MDS projects on GitHub.
My master's thesis: an intelligent agent for Spanish accounting regulation. Rather than relying on probabilistic LLM answers, I'm turning the legislation from the BOE (Boletín Oficial del Estado) into a formal knowledge graph — an OWL/RDF ontology — then linking invoices to it and answering accounting questions through semantic inference over the graph. The guiding principle is correctness over fluency: the ontology has to be logically verifiable, so the original regulation can be reconstructed from it.
HomeInfra
The Terraform-managed home lab that runs this very website. It provisions a k3s Kubernetes cluster on Proxmox VE — with Traefik ingress, Let's Encrypt TLS, Tailscale access and TrueNAS-backed NFS storage — where every app deploys itself from its own repository through in-cluster CI runners. Everything you're looking at is served from it.
A tool to split tickets and shared expenses across a group — think Tricount, but heading towards scanning a receipt and letting an LLM pull out the line items for you. It splits equally, by parts or by exact amounts (per ticket or line by line), tracks balances and suggests the minimal set of payments to settle up — with all the money math done in integer cents so splits always add back to the exact total. Built with Next.js and Prisma on PostgreSQL.
A small app to keep track of the groceries I have at home, so I always know what's running low. Built with Next.js and Prisma on PostgreSQL and self-hosted on my home cluster.
A fast, keyboard-driven terminal app for managing grading projects and tasks. I built it while teaching a few classes at the UPC to keep track of my students' grades — a TUI written in Python with Textual, backed by SQLite.
The very site you're looking at. A dark-themed Next.js portfolio with a full-screen landing and scroll-driven animations, built with Tailwind and framer-motion — and, like everything else here, self-hosted on my home cluster, deploying itself on every push.
Live on the web
Web pages
As well as working on personal projects, I love to create small web pages that solve everyday problems, or that are just fun to use. Here are some of the web pages I've created:
I needed a way to collaboratively track all the flats that two colleagues and I found over the internet, to rent during the 2025-2026 academic year. So, I quickly spun up a web application using NextJS + Prisma to expose and interact with a PostgreSQL database, which I hosted on my Proxmox home lab. The page is still in a very early stage, so make sure to keep an eye on its GitHub repository if you want to see the progress.
Create teams of as many people as you want with as many constrains as you need. Maybe you want to distribute the seats in the work meeting, or maybe you want to create a balanced team for a sports game. Teamificator will allow you to ensure that you sit next to your crush and you don't have to talk to your boss. This web page is currently being developed. You can keep an eye on its GitHub repository if you want to see the progress.
I have a Minecraft server with my friends, and I wanted to create a page where we could see the server status, the players online, and manage the backups. I created a simple web page that allows me to do just that using flask.
I visited London a few years ago, and loved it! The problem is that I have a terrible memory, and wanted to visit as many iconic places as possible. The solution: create a map with all the places I wanted to visit.
This website — the one you're on right now (or its .dev twin). I built it to showcase everything above: a dark, animated Next.js portfolio, served straight from my home Kubernetes cluster.Curiosity-driven
Studies
I tend to have a critical eye on the world around me, so here are some small studies I've done on different topics:
Some time ago, I saw some kid on Tik Tok saying that there was no point in studying anything because doing so wouldn't increase your income. I thought that was a pretty bold statement, so I did what any reasonable person would do: I went to Kaggle to find a dataset that would help me prove him wrong.
Career so far
Experience
AI Engineer at Kabilio
Kabilio is a two-year-old startup that has already grown to around 70 people, building the next generation of AI-powered Spanish accounting software, a platform designed from the ground up around AI decision-making, automation and agents. We run the full spectrum of AI: classical ML models, LLMs for classification and summarization, RAG, rule mining, recommendation systems, fully autonomous agents and chatbots. I joined as the company's second AI engineer and I'm a key part of the AI team, now four AI engineers working alongside 20+ full-stack, DevOps and data engineers.
Research Engineer at UPC
For around six months I worked in a research group at the Universitat Politècnica de Catalunya (UPC), building a Data Space for the Catalan agricultural sector. I designed and developed a complex orchestrator that managed all the data connectors deployed across a Kubernetes cluster.
AI Engineer at Awaait
Awaait Artificial Intelligence builds fully-distributed, on-premise computer vision for smart, holistic fare-evasion detection. We ran dozens of servers inside stations of the Ferrocarrils de la Generalitat de Catalunya (FGC) and the New York City subway, detecting fare evasion in real time — people jumping over turnstiles or slipping in through emergency exits — and raising live alerts so station security agents could stop them in the act. We also provided the transport authorities with aggregated metrics to estimate the impact of fare evasion across their stations.