Curriculum Vitae
Frontier AI evaluation, harmful manipulation and epistemic risk, and how public institutions actually adopt these systems.
Cambridge ERA:AI Research Fellow · Senior AI Advisor, Office of the PM of Spain, 2025–26 · Founder, Saturdays.AI
Which model behaviors count as manipulation rather than persuasion, how persuasive and agentic systems change what people come to believe and decide, and what evaluations would catch that before deployment. The focus of my Cambridge ERA:AI research program on harmful manipulation and election security.
The research program →Inside the Office of the Prime Minister of Spain, 2025–26: how a state adopts AI — agentic applications, shared infrastructure, risk assessments and deployment safeguards, and the capacity to run and question them. Three initiatives I led there were selected, following OECD review, for the OECD.AI Policy Observatory.
Lessons from adopting AI in government →Building the instruments the arguments depend on — FronteraEval, an open catalog of frontier AI evaluations and their limits; reference datasets; audits such as a FORTRESS run on a frontier model 48 hours after release — on the premise that governance advice should stay close to how these systems actually behave.
Explore FronteraEval →An international non-profit for AI education, grown mostly through volunteer-led local chapters: 12 countries, 30K+ alumni, and 500+ AI4Good projects built by participants.
Democratizing AI education →Jul 2026 – present
Cambridge ERA:AI Fellowship · Cambridge, UK
Research fellowship on frontier AI safety and governance. Designed and ran a four-part research program on harmful manipulation and election security, connecting how manipulation operates, what evaluations show, real-world cases, and policy interventions. Its working paper, The Capability–Deployment–Effect Gap, separates model capability, system deployment, and demonstrated effects, with an explicit ceiling on what each claim can support. Audited 3,360 evaluation records and mapped 68 control families across 118 policy and technical implementations, published as source-linked artifacts built for scrutiny, replication, and policy use.
Mar 2025 – Oct 2026
Office of the Prime Minister of Spain · PresidencIA
Advising on AI strategy, safety, and state capacity: translating model capabilities and failure modes into evaluation criteria, procurement requirements, risk assessments, and deployment safeguards, and working across technical, policy, legal, and security teams on adoption and governance decisions. The work spans PresidencIA — the cabinet-wide AI adoption effort organized around community, applications, and infrastructure, which has mobilized hundreds of public officials — and agentic AI applications from an AI-scientist assistant to crisis response. Led, designed and built three of these initiatives — PresidencIA, the ServetIA research platform, and CiudadanIA for citizen correspondence — all selected following OECD review for publication in the OECD.AI Policy Observatory. Sitting inside the institution is what kept the research grounded in what these systems actually do once deployed.
2026 – present
In the AI governance program of an independent Brussels think tank: research, dialogue, and training on what AI means for regulation, global security, and equity — the European and international side of the questions raised by the evaluation research and the work inside government.
Fall 2026
SPAR · Supervised Program for Alignment Research
Mentored AI safety research project on the boundary between helpful persuasion and harmful manipulation: what separates them in model behavior, and how an evaluation could tell them apart — the line the ERA:AI work needs drawn before it can be measured.
Fall 2026
Talos Network · Policy Leaders Programme
Policy fellowship on AI governance, run alongside the research: the same questions about frontier systems, carried into the places where the rules for them are written.
2018 – present
Saturdays.AI
Founded the organization and helped it grow to 12 countries, 30K+ alumni and 500+ AI4Good projects — most of that carried by volunteer-led local chapters rather than by me. Designed the AI Saturdays bootcamp and dozens of AI courses. Ran early LLM-for-social-impact projects with dozens of engineers.
2021 – 2025
Selected Consulting Engagements
The deployment side of the same question — what frontier systems do once they are load-bearing inside an organization.
Professional Services
Manager. Led development of an Agent OS platform.
Infrastructure
Tech Lead on a GenAI platform for a large infrastructure company; Architect on an AI agent deployment in finance.
Defense & Aerospace
Consultant. Helped define AI use cases for national defense.
Energy
Senior Engineer. Developed RAG systems for industrial engineering.
2019 – 2025
Epitech
Taught machine learning to graduate students, alongside the hands-on AI work of those years.
2023 – 2024
Tauniqo.AI
Led a team of engineers building an LLM-driven platform for training and talent selection. Built the MVP the company took into a seed round, and onboarded the first enterprise clients.
2019 – 2020
Akademy.AI
Led technology and operations for an AI bootcamp, later acquired by a YC startup. Trained AI engineers, with a success rate above 95%.
2013 – 2014
Microsoft
Advised on cloud product-market fit in frontier markets across Benin, France, and Rwanda. Co-led a team of 11 consultants spread across 4 continents on technology adoption initiatives.
Harmful Manipulation & Election Security
Working research on how capability evidence, deployment conditions, observed incidents, and intervention evidence should change policy conclusions about AI manipulation — stating explicitly what the evidence demonstrates and which downstream effects it does not establish.
Cambridge ERA:AI research program · source-linked
Evaluation Catalog
Open catalog for finding a frontier AI evaluation, following it back to its original source, and reading its limits. Every record resolves through a versioned registry that keeps the evaluation's developer, its paper, and its registry host apart — and says so when a source is unresolved, rather than supplying a citation that does not exist.
Open source · open data, refreshed weekly
Evaluation Infrastructure
An open synthetic reference population for Spain, published as a benchmark substrate for testing whether AI systems work for the country they serve — with a datasheet, an evaluation report, and a privacy and disclosure note.
Open dataset · Hugging Face
Evaluation Tooling
A local-first workbench for summarization and evaluation: compare runs across models, score outputs against weighted criteria, and mask personal data locally before anything reaches a provider.
Open source · desktop and BYOK demo
AI Policy
Independent, source-backed guide to Spain's AI governance — AI Act timelines, AESIA, ALIA, sovereign compute, and frontier autonomy, with links to primary sources for policymakers, researchers, and journalists.
Independent · primary sources linked
Civic AI
Democratic intelligence platform combining AI tools, analysis, and publications to help citizens vote with more information and criteria — built for Colombia's 2026 presidential elections.
Colombia 2026 elections
Youth Mentorship
Non-profit connecting industry professionals with student teams to build social-impact technology projects over 12–16 week mentorship cycles.
45+ projects · 15K+ people impacted
Healthcare AI
Avatar technology that builds representations of loved ones, intended to support people living with Alzheimer's, dementia, and cognitive decline.
Built by a Saturdays.AI team
Social Impact
AI chatbot helping refugees arriving in Spain navigate the reception system by answering questions about resources, services, and integration support.
Refugee assistance
EdTech
Educational entertainment platform that aims to turn screen time into active learning by connecting children and young adults with influencers and curated educational content.
Early stage
AI for Educators
AI skills laboratory training teachers across educational centers in using AI tools for classroom innovation, from primary through vocational education.
280+ teachers · 11 centers
International Fellow in Deep Learning
Master of Business Administration (MBA)
Diploma in Consulting and International Business
Licenciatura in Economics
Advanced Vocational Diploma in Computer Science (Técnico Superior)