DEMIURGE builds Generative Immersive Virtual Models that create fully immersive training environments from a description — enriched with real-time physics and haptic feedback, for humans and robots alike.
Explore the vision ↓Dynamic Environment Modeling for Immersive Universal Reality Generation Engine
Vision
Just as large language models generate text, ideas, and code from a simple prompt, DEMIURGE's Generative Immersive Virtual Models (GIVM) generate complete virtual training environments from a description or a learning objective. Environments enriched with haptic and sensory feedback that make experience indistinguishable from reality — for trainees, surgeons, and autonomous robots.
"We imagine a future where you describe a complex surgical procedure and our model creates — in seconds — an immersive, adaptive environment to train it."
The Problem
The gold standard for realism, but costly, risky to patients, and entirely dependent on case availability. Cannot scale.
HIGH RISK · NOT SCALABLEGood haptic realism, but limited scenario variability, high update costs, and only partially reusable. Static by nature.
LIMITED · EXPENSIVEScalable and affordable, but low physical accuracy, no generative capability, and no automatic scenario customisation.
LOW FIDELITY · STATICTechnology
Every component is a research frontier. Together they form an integration no competitor currently offers.
01
Automatic creation of new immersive scenarios from text descriptions. Real-time simulation of complex physical phenomena — tissue deformation, fluid dynamics — too costly for traditional engines.
02
Faithful real-time simulation of soft tissues, surgical instruments, and their interactions. Accurate enough to train both human surgeons and robotic systems.
03
Realistic perception of texture, stiffness, and reaction forces. Haptic data feeds back into model training — tactile embodiment that transforms sensation into material knowledge for adaptive manipulation.
04
Augmented, virtual, extended, and mixed reality combined for multisensory, intuitive interaction. Users move fluidly between modes without breaking immersion.
05
Privacy-preserving data sharing by design, using advanced cryptographic protocols. Enables sensitive clinical data to be used for model training without exposure across institutions.
06
Mechanisms designed to encourage cooperation between heterogeneous institutions — hospitals, universities, research centres — making data sharing mutually beneficial and sustainable.
Competitive Positioning
Platforms like NVIDIA Isaac Sim and Unreal Engine excel at simulation. Google DeepMind's Project Genie and NVIDIA Cosmos lead in AI generation. DEMIURGE is the only platform that integrates both — and adds haptics, secure data sharing, and cooperative incentives.
| Platform | AI Generation | Physics Simulation | Haptic Integration | Secure Data Sharing | Scalable Scenarios |
|---|---|---|---|---|---|
| DEMIURGE | ✓ | ✓ | ✓ | ✓ | ✓ |
| NVIDIA Isaac Sim / Unity / Unreal | ✗ | ✓ | ~ | ✗ | ~ |
| Project Genie / NVIDIA Cosmos | ✓ | ✗ | ✗ | ✗ | ~ |
| Osso VR / PrecisionOS / FundamentalVR | ✗ | ~ | ~ | ✗ | ~ |
| Physical simulators (Simbionix, Laerdal) | ✗ | ~ | ✓ | ✗ | ✗ |
Roadmap
PHASE 1 · MONTHS 1–6
Integrate first generative features into the existing VR prototype. Validate business model hypotheses. Conduct patentability study. Establish startup structure and roles. Funded by Sapienza grant.
PHASE 2 · MONTHS 7–12
Deliver the Simulator Core (VR system + AV + training) to the first early adopter — Policlinico Umberto I or Gemelli. Validate the technical support and clinical training model. Activate recurring service agreements.
PHASE 3 · MONTHS 13–24
Activate annual service and update streams (~€55K per client). Expand into synthetic dataset sales for surgical robotics R&D, using the video and debriefing infrastructure already installed at client sites.
PHASE 4 · BEYOND 24 MONTHS
Scale the GIVM engine to biology, chemistry, industrial manufacturing, and complex robotics. Reduce hardware dependency as the generative model matures. Position DEMIURGE at the frontier of physical agentic AI.
Team
Internationally recognised expertise in AI, physics simulation, cryptography, and game theory — a combination essential for this challenge and rare in a single group.
Strategy & Tech Transfer
Startup experience; WSENSE co-founder. Bridges academic research and market.
Game Theory & Institutions
Data sharing incentives and institutional cooperation. Industry experience with Google.
Physics Modeling Lead
Built the first DEMIURGE demonstrator. Expert in dynamic systems and medical robotics.
Generative AI Lead
Automatic creation of immersive modules. Industry collaboration with Google and Meta.
Cybersecurity & Cryptography
Designs secure, confidential protocols for multi-institution data sharing.
Physics & Robotics · PhD 2023
Physics modeling and immersive systems applied to robotic surgery.
Generative AI · PhD 2024
FAIR Research Fellow. Implements complex physics simulations in real time.
Topological ML · PhD 2022
Topological machine learning and physics modeling applied to medicine.
Decentralised Infra · PhD 2024
Distributed ledgers and advanced cryptographic tools for data governance.
External Collaborators
Gemelli IRCCS · IHU-Strasbourg
Clinical expertise and validation. VBA&CAI research facility lead.
Sapienza · Radiology
Interventional radiology and breast pathology. EUSOBI Executive Board.
Sapienza · Simulation Lab
Director of the Sapienza Skill Lab. Clinical simulation design expertise.
Get Involved
We are looking for early adopters, research partners, and investors who want to be part of building the generative engine for immersive virtual worlds.