Supercomputing (HPC)
We run our most demanding models and simulations on European supercomputers.
Back to home
Technology
Technology and research
For those who want to know what is under the hood: our technology, our publications and how we protect data.
Technology
What’s under the hood.
Sovereign AI
Our models run on our own hardware in the EU, not on a public cloud. Your data never trains third-party models.
Custom model training
We build every kind of AI model: from machine learning and deep neural networks to physics-based models, computer vision and large language models. For each problem we choose the architecture that fits it best.
Quantum computing
Quantum algorithms and simulations for chemistry, new materials and other calculations that classical computers cannot handle well.
Advanced data analytics
Predictive models of behaviour, preferences and demand. Not simple statistics, but models that learn how people and markets really behave, with results in live dashboards.
AR, VR & 3D
Unity, augmented reality, LiDAR spatial mapping and large-format 3D printing.
Science
Backed by science, not marketing.
We train our models on supercomputers, verify them with validation tests following scientific methods, and publish the results. No wrapper around someone else’s API pretending to be AI.
arXiv:2608.13069 · 2026
Behavioral Modification Boundaries of Open-Weight Large Language Models Under Direct Preference Optimization
How far the behaviour and personality of open-weight language models can be changed with DPO: learning curves, base vs. instruct models, cross-lingual transfer and personality stress tests.
arXiv:2609.20439 · 2026
Representation Inequivalence in Open-Shell ADAPT-VQE: Projected Optimization versus Full-Space Implementation
When a projected quantum-chemistry optimisation and its full-space implementation stop being the same thing, with diagnostics on copper, NO and OH systems and a validated Qiskit circuit.
Trust
Built on European law.
Every solution is designed from the start to comply with these regulations.
EU AI Act, Regulation (EU) 2024/1689
Risk classification of each system (Art. 6 and Annex III), transparency obligations (Art. 50), and for high-risk systems technical documentation, record-keeping and human oversight (Art. 11, 12 and 14).
GDPR, Regulation (EU) 2016/679
Lawfulness and data minimisation (Art. 5 and 6), data protection by design and by default (Art. 25) and security of processing (Art. 32).
Data processing agreements
For client solutions we act as processor under a written agreement (GDPR, Art. 28) and keep a record of processing activities (Art. 30).
Spanish data protection law
Organic Law 3/2018 (LOPDGDD). Supervisory authority: the Spanish Data Protection Agency (AEPD).
Data sovereignty
Models are trained and run on our own hardware in the EU. Client data never trains third-party models and is not transferred outside the EU (GDPR, Chapter V).
This website
No cookies and no tracking (ePrivacy Directive 2002/58/EC, Art. 5(3); Spanish LSSI, Art. 22.2), no external fonts or scripts.
Written about us
REDCIDE · Success storiesThe Safe Haven: artificial intelligence, supercomputing and augmented reality at the service of bioclimatic construction in TenerifeIY Magazine · EntrepreneursThe Safe Haven: innovation in sustainable construction by Modelos Inteligencia Artificial S.L. with MentorDayMentorDay · Press releasesThe Safe Haven, a RED CIDE and MentorDay success storyContact
Want a solution like this for your company?
Describe your idea, even if it is still taking shape. We will reply with a concrete proposal.
- lucy@thesafehaven.ai
- Tel.
- +421 907 536 490
- Address
- C/ Tajinaste, 54, Pta. G4, 38620 San Miguel de Abona, Tenerife, España