Software Craftsman · Cyberpunk · CTO
Nearly two decades designing system architectures, leading AI teams, and building digital products at global scale.
New podcast Catalan-language podcast on AI Humanist and ethical lens
With Marçal Font & Xavier Vinaixa
The first Catalan-language podcast that looks at artificial intelligence through humanism, ethics and philosophy. No apocalypse, a positive outlook, and Catalan Sign Language (LSC).
Active investigation 12 outlets (ES · UY · AR · AD) May–June 2026
Joint investigation Xavier Vinaixa & Marçal Font Espí · Llibreria Fènix (Badalona)
How a mysterious company buys second-hand books from bookstores across half the world to train AI models — then physically destroys them. Original investigation published on Substack and picked up by elDiario.es, Cadena SER, VIA Empresa and Montevideo Portal.
Viral reel on @3catinfo (TV3) 700k+ viewsOn RTVE's «Cafè d'idees»
Upcoming event +RAIN Film Fest 2026 16 June 2026
Presentation of the Signària project and a roundtable in the MEDIA+RAIN section on the role of AI in journalism and communication, organised by UPF's Department of Communication.
Roundtable with Veronika Solopova (DFKI · TU Berlin), Rob Lang (Reuters), Xavier Vinaixa and David Quiñonero (ARA) · moderated by Roger Cassany
+RAIN Film Festival — Europe's first festival of AI-made cinema (UPF + Sónar) · Room 55.309, Poblenou Campus · 14-17 June 2026
Latest publication LREC2026 16 May 2026
Authors Gemma Barberà · Inés Broto Clemente · Xavier Vinaixa Roselló · Roger Cassany Viladomat
12th Workshop on the Representation and Processing of Sign Languages: Language in Motion · Palma de Mallorca · ISBN 978-2-493814-82-1 · CC BY-NC 4.0
Software craftsman and artificial intelligence specialist with a humanist and ethical vision. Nearly two decades creating and scaling digital products — from mobile apps with 55 million downloads to immersive virtual reality projectors for hospitals. All my work, both technical and in science communication, is grounded in a firm commitment to science, rigor and honesty, and in the fight against anti-intellectualism and pseudoscience.
Currently leading AI research at Sorensen.ai, where I serve as technical lead of Signària — an EU-funded project (Next Generation EU) to create personalized radio sonic identity for deaf sign language users, led by UPF. The project received an Honorable Mention at the 26th Ràdio Associació de Catalunya Awards (2026).
Public television contributor (Badalona 360). Co-creator of the La Màquina de Llull podcast — the first Catalan-language podcast on AI with a humanist, ethical and philosophical lens. Creator of python-toon — the reference Python library for TOON with 2M+ installs — and collaborator on TOON Format with its own Wikipedia page. President of Opendata.cat and creator of the Opendata.cat MCP server — connecting any AI with 2,850+ public datasets. Author of academic papers on LLM hallucination, literary style fine-tuning and personalized sonic identity for deaf signers (XVII CILCS 2025).
Ethics & Humanism
Transparent, explainable models oriented toward bias mitigation. Human-centered technology. A commitment to science, rigor and honesty, and to the fight against anti-intellectualism.
Art & Music
Bass player in LLAMP (EP Claredat 2023), The Singletons (200K+ listens), Elite Funk (EP 2012), DeMortimers and Ateneuats Band (jazz). Creator of a musical instrument built with floppy drives and MIDI.Co-founder of LOVYU — immersive AI art installation (600+ visitors). Lighting consultant for Reflexes by ESDi at Llum BCN. Collaborator on the Estimat Alvar exhibition (tribute to Álvar Suñol).Co-creator and developer of FestaFoto — first rapid mobile photography contest in Tiana (2011). First remote immersive concert without headsets at Sónar during lockdown, using BROOMX's live 360 system.
Human Factor
SUP competitor at CN Bétulo. 3rd in Catalonia indoor rowing 2023 with REM Premià. Co-founder of 8rems.com — the reference platform for Catalan llagut leisure rowing with 10+ years of activity.Gold Medal at BCN Beer Fest with Brew by Code. President of ABAT (Craft beer association of Tiana). Founder of the Cervesers.cat cooperative.Cooking channel created during COVID (200+ subscribers, videos with 30K+ views). President of Opendata.cat (Open data advocacy association).Devil dancer for 20 years in the Trempera Nostramo fire-running troupe (Tiana).
Outreach
Monthly contributor on TV Badalona 360 and Radio Ciutat de Badalona. Catalan-language newsletter on AI and humanism. xaviviro.substack.com
// Sorensen.ai
As CTO, I lead AI research applied to media, culture, and the arts. The goal: turning AI into accessible tools — summaries, easy-to-read versions, sign language translations, and real-time subtitles. Creator of Larsson.ai — AI assistant for journalists and content creators.
Project funded by Next Generation EU and led by UPF to create personalized radio sonic identity for deaf sign language users. A unique, personal synthetic voice — not generic — that represents each person's identity in radio and podcasting. Hybrid architecture combining symbolic NLP with Neural TTS and LSC processing.
AI assistant for journalists and content creators. Content creation, automated podcasts with voice synthesis, multilingual video dubbing, SEO/AEO/SMO optimization and multilingual transcription. Private servers, efficient models and native Catalan, Basque and Galician support.
→ Larsson.aiEvery project with verifiable references.
LLM · Political Bias · Jensen-Shannon
Author · Sorensen.ai · 2026
Empirical study measuring whether an LLM changes its political answers depending on whether you speak Catalan or Spanish. Ten models from Google, Anthropic, OpenAI, Groq and DeepSeek tested against 21 survey items with real data from the CEO and the CIS. Survey-grounded method (MENAValues), Jensen-Shannon distance and bootstrap confidence intervals. Result: Gemini shifts the most (JSD 0.556 and 0.534) while DeepSeek answers almost identically in both languages (0.037 net). The shift concentrates on independence, national identity and institutional trust; the left-right axis and the economy are the most stable. Open code (MIT) and data on GitHub and Hugging Face.
LLM · Embeddings · KNN
Author · Sorensen.ai · 2026
Empirical study showing that even when you train an LLM in Catalan, the concepts still come in English. Eleven open-weight models tested — Gemma, Qwen, Mistral, DeepSeek, and even the Catalan-native Salamandra and ALIA models from BSC. Result: 10 of 11 models, with no context, place coca next to cocaine and Coca-Cola in embedding space. The study proposes a taxonomy of three lexical-cultural collision modes (F1 lexical-cluster collision, F2 Anglo narrative drift, F3 polysemic dominance) and shows that the most effective intervention is not retraining, but providing Catalan cultural context at inference time. Code (MIT), data (CC-BY 4.0) and methodology open on GitHub.
MCP · AI · Open Data
Creator · President of Opendata.cat · 2026
The weapon against black boxes: an open MCP (Model Context Protocol) server that connects any language model — from a local open-weights LLM to ChatGPT, Claude or Gemini — with 2,850+ datasets from 1,000+ municipalities and state agencies (Generalitat, Barcelona City Council, Barcelona Provincial Council, AOC Consortium, INE, REE, CNMC, Rodalies, FGC…). Built so you can interrogate reality directly without ideological filters: the cheapest gas station, the best hour to run a washing machine, reservoir levels during a drought, real-time Rodalies or FGC delays. Every answer carries a traceability chain — dataset, last update, originating institution — to dismantle hallucinations. Inspired by the French datagouv-mcp, but designed to understand the Catalan territory. npm package compatible with Claude Desktop, Claude Code, VS Code, and Cursor. An Opendata.cat project (Open Data Promotion Association).
AI · LLM · Data Format
Creator of python-toon · TOON Collaborator · 2025
Creator of python-toon, the reference Python library for TOON (Token-Oriented Object Notation) with 2M+ installs on PyPI — open data format reducing LLM tokens by 30-60%. De facto standard for efficient data serialisation in LLM environments. Collaborator on the official project. Wikipedia page, coverage in InfoQ, freeCodeCamp, Analytics Vidhya, and Hacker News.
LLM · Fine-tuning · Catalan
Creator · 2023–2024
FLAMA — first small model to speak Catalan (Llama 3B fine-tune, 2023). SUGARRA — first small model in Basque (Llama 3B fine-tune, 2024). Creator of the FEDERICO-GARCIA-LORCA dataset (songs, poems, and ballads). LoRAs published on CivitAI, including CyberNeon FLUX (1.2K favs).
GANs · Diffusion · Python
Creator · Open Source
One-click multiple face swap in video. Built on GANs and diffusion models. GitHub: Pair Extraordinaire, Pull Shark.
VR · MedTech · Hardware
Co-founder & CTO
First immersive VR projector for hospitals. Real-time projection mapping, spatial sensors, geometric correction to prevent cybersickness. Selected for the Abbey Road Red incubator (London) to explore immersive audio and spatial audio techniques.
IoT · Open Source · Sensors
Architect & Developer · Fab Lab Barcelona (IAAC) · Since 2012
Designed the architecture and built the first version of the platform and the first iOS app. Open-source citizen environmental sensor platform deployed in Barcelona, Amsterdam, and Manchester with partners like Cisco and Intel. Referenced on DATA.GOV (US). Deployed a sensor network in Tiana (2020-2021).
Open Data · Data Visualisation · Tailwind CSS
Creator · Opendata.cat · 2026
Open data platform with 1M+ vehicle registrations across Catalonia (2020–2026). Interactive visualisations by municipality, manufacturer, fuel type, CO₂ emissions, and electrification trends. Data sourced from DGT and the Catalan Government Transparency Portal. An Opendata.cat project (Open Data Promotion Association).
Open Data · Data Visualisation · Chart.js
Creator · Opendata.cat · 2026
Open data platform mapping Catalonia's bike lane network. 2,370 km mapped across 327 municipalities. Interactive map, rankings by kilometres and per capita, municipal comparisons. Data from OpenStreetMap, Open Data BCN, and Generalitat de Catalunya. An Opendata.cat project.
AI Art · Generative · Installation
Co-founder · 2024
Immersive AI art installation simulating Mario Tarragó's painting style to generate 'soul portraits'. 600+ visitors, ~360 portraits generated.
Music Tech · Hardware · MIDI
Creator
Musical instrument built with floppy disk drives controlled via MIDI. The drives generate musical notes through stepper motor head movements.
Music Tech · Hardware · IoT
Lead Developer · 2014
World's first digital handpan. Electronic instrument with multi-gesture sensors, Bluetooth, and MIDI. €348K raised on Kickstarter with 680 backers.
AI · Music · NLP
Creator · 2020
AI-powered lyrics aligner for karaoke. YouTube channel with 900+ subscribers and videos with 40K+ views.
Streaming · Platform · COVID
Co-creator · 2020
Streaming platform for ceremonies and funerals during COVID, integrated into Complementos Funerarios. Enabled families to remotely attend live ceremonies when restrictions prevented physical presence.
VR · Open Source · Art
Developer · Gabriel Lecup / Hangar BCN
Open source VR app to explore altered states of consciousness with eyes closed. Combines flickering lights and binaural beats to induce lucid dreams, trances, and hallucinations through brainwave synchronisation (FFR). Produced with the collaboration of Hangar — Interaction Lab.
Sports · Community · Platform
Co-founder · Since 2014
Reference web platform for promoting and organising fixed-seat rowing events with Catalan llagut boats in leisure format. Meeting point for rowers, organisers, and enthusiasts with 10+ years of activity.
WhisperX · Axolotl · llama.cpp
Contributor
WhisperX (forced alignment, diarization). Axolotl (LoRA training). llama-cpp-python (local quantized model execution).
Three decades of products shipped to production — with metrics, reference press coverage, and verifiable deployments.
LREC 2026 · RAC Award 2026
EU Next Generation project (UPF + Sorensen.ai). Pioneering tool released to production in March 2026. Paper at LREC2026 on synthetic and 3D training data for LSC plus Honour Mention at the 26th Ràdio Associació de Catalunya Awards. Coverage: El País, ARA, TV3, betevé, VilaWeb.
Hospitals + Abbey Road
Co-founded the startup that built the first headset-free VR projector for hospital environments (geriatrics, oncology). Selected for the Abbey Road Red incubation programme. Coverage: El Mundo, La Vanguardia, El Periódico, Music Week.
2M+ installs
Creator of the reference Python library for the TOON format (2M+ installs on PyPI) and official collaborator on the project. Coverage on InfoQ, freeCodeCamp, Hacker News, Analytics Vidhya, and a dedicated Wikipedia page.
55M+ downloads
Founded and led the technical work of the mobile studio that shipped over 40 apps with more than 55 million downloads worldwide, predating today's monetisation models.
Leading AI research for accessibility. Signària, real-time computer vision, NLP, sign language translation.
Immersive VR projector for MedTech. Projection mapping, geometric correction, spatial sensors.
Connectivity services management and scalability platform.
40+ apps, 55M+ downloads. Distributed systems, massive concurrency, payment gateways.
First open data portal of a Spanish municipality. BDN Bústia: first app for citizens to report public road incidents to city council with tracking.
Newsletter
Exploring how to maintain human purpose. AI critique, algorithmic ethics, techno-feudalism, death of the author.
Subscribe to the newsletter
xaviviro.substack.comPodcast · Guest
Guest on the Catalan podcast Com dir-t'ho, produced by the Escola de Formació Guillem Agulló (Òmnium Cultural) and hosted by Jana Jubert and Laura Grau, discussing the technical and economic dimensions of artificial intelligence: what language models can actually do, their limits and risks, and how AI is reshaping the job market. With economist Enric Vila.
04/06/2026 · 39 min
TV · Sign Language
TV3 report on Signària: a pioneering tool created by UPF and Sorensen.ai that enables deaf signers to participate in radio with a personalized synthetic voice.
04/04/2026 · 2 min 15 s
Speech · Video
Collecting the Honor Mention at the 26th Ràdio Associació de Catalunya Awards for Signària, with a radio identity synthesized with AI. «Now, the radio signs too.»
21/05/2026 · 1 min 32 s
Signària pageRadio
"A true problem solver" — From creating an app to win at Apalabrados to developing an immersive VR projector with entertainment, educational, and medical applications.
15/03/2022 · 23 min
Radio
Masterclass on artificial intelligence on L'Entrellat, Ràdio 4 (RTVE). Conversation with Rosa Pera about the present and future of AI.
RTVE · Ràdio 4
Apps · Coverage
Multiplayer trivia with 60,000+ questions in 22+ categories, including the infamous "Corruption/Spain". 7,000+ downloads at launch. Co-created with Sergi Solé. Covered by Voz Libre, Applicantes, 3Cat/TV3.
Talks & Workshops
Participation in forums, conferences and workshops on AI, open data, tech ethics and social impact.
Social Media
Short-form reflections on tech news and AI from a critical, humanist perspective. On the @wearesorensen Instagram account.
Public TV
Regular contributor on public television explaining AI, ChatGPT, foundational models, and tech ethics to the general public.
YouTube PlaylistPress
Articles in reference publications — from TOON Format coverage in InfoQ and freeCodeCamp, to interviews on Catalunya Ràdio, El Español, and VIA Empresa.
Available for advisory, keynotes, technical collaborations, and social impact projects.
