Skip to main content
NVIDIA AI Podcast

How Dassault Systèmes Is Building AI That Understands Physics - Ep. 296

23 min episode · 2 min read
·
Nicolas Saricier

Episode

23 min

Read time

2 min

Topics

Relationships, Fundraising & VC, Leadership

AI-Generated Summary

Key Takeaways

  • Industry World Models vs. Generative AI: Standard generative AI predicts outcomes by observing patterns — it can predict a plane will fly but cannot explain why. Dassault's industry world models embed actual physics, chemistry, material science, and engineering laws directly into AI reasoning, ensuring outputs are scientifically valid rather than statistically plausible. Engineers should evaluate AI tools by whether they encode domain laws, not just training data.
  • Virtual Companion Architecture: Dassault deploys three specialized AI agents — Aura (business), Leo (engineering), Marie (science) — each reasoning through industry world models rather than general LLMs. Leo, for example, takes a 3D scan or 2D drawing, runs physics and kinematics analysis, and returns a manufacturable, optimized design. Organizations building AI workflows should consider role-specific agents over generalist models for regulated domains.
  • NVIDIA Integration Delivers Measurable Gains: Integrating NVIDIA NIM models improved document ingestion throughput by 30%, while Nemotron reasoning models improved performance for Aura, Leo, and Marie by 20% without task-specific optimization. Teams deploying agentic systems should benchmark NIM deployments against existing pipelines — containerized NIM models reduce integration friction significantly across Kubernetes infrastructure.
  • IP Lifecycle Management as Trust Infrastructure: Dassault enforces full traceability of every AI interaction through a system called IP Lifecycle Management (IPLM), logging which workflows, models, and processes modified any content. Industrial AI deployments require this audit layer — without traceable lineage, human accountability breaks down in regulated environments. Build traceability into agentic architecture from day one, not as an afterthought.
  • Hybrid Model Strategy for Sovereign Compliance: Dassault combines proprietary models with frontier models from NVIDIA (Nemotron via NIM) and Mistral, selecting partners based on performance and sovereignty constraints. For global industrial customers in regulated sectors, model selection must account for regional data regulations and auditability requirements. Open standards like MCP and agent-to-agent protocols enable cross-system agentic choreography without vendor lock-in.

What It Covers

Dassault Systèmes VP Nicolas Saricier explains how the company is shifting from a SaaS platform to an "agent as a service" model, deploying physics-grounded AI virtual companions named Aura, Leo, and Marie to serve 45 million engineers and scientists across regulated industries worldwide.

Key Questions Answered

  • Industry World Models vs. Generative AI: Standard generative AI predicts outcomes by observing patterns — it can predict a plane will fly but cannot explain why. Dassault's industry world models embed actual physics, chemistry, material science, and engineering laws directly into AI reasoning, ensuring outputs are scientifically valid rather than statistically plausible. Engineers should evaluate AI tools by whether they encode domain laws, not just training data.
  • Virtual Companion Architecture: Dassault deploys three specialized AI agents — Aura (business), Leo (engineering), Marie (science) — each reasoning through industry world models rather than general LLMs. Leo, for example, takes a 3D scan or 2D drawing, runs physics and kinematics analysis, and returns a manufacturable, optimized design. Organizations building AI workflows should consider role-specific agents over generalist models for regulated domains.
  • NVIDIA Integration Delivers Measurable Gains: Integrating NVIDIA NIM models improved document ingestion throughput by 30%, while Nemotron reasoning models improved performance for Aura, Leo, and Marie by 20% without task-specific optimization. Teams deploying agentic systems should benchmark NIM deployments against existing pipelines — containerized NIM models reduce integration friction significantly across Kubernetes infrastructure.
  • IP Lifecycle Management as Trust Infrastructure: Dassault enforces full traceability of every AI interaction through a system called IP Lifecycle Management (IPLM), logging which workflows, models, and processes modified any content. Industrial AI deployments require this audit layer — without traceable lineage, human accountability breaks down in regulated environments. Build traceability into agentic architecture from day one, not as an afterthought.
  • Hybrid Model Strategy for Sovereign Compliance: Dassault combines proprietary models with frontier models from NVIDIA (Nemotron via NIM) and Mistral, selecting partners based on performance and sovereignty constraints. For global industrial customers in regulated sectors, model selection must account for regional data regulations and auditability requirements. Open standards like MCP and agent-to-agent protocols enable cross-system agentic choreography without vendor lock-in.

Notable Moment

Saricier describes a customer — Nayar — that physically disassembles aircraft and uses Leo to reconstruct thousands of parts as 3D digital models without access to original design files. This reverse-engineering workflow, previously manual and time-intensive, now runs automatically from scans and drawings.

Know someone who'd find this useful?

Episode Transcript

The the agents, can use a virtual twin as a gym, to train themselves. So they can run, in fact, millions of simulations or design experimentation and present, present to you, to the to the human, to the engineer, the proven solution. Welcome to the NVIDIA AI podcast. I'm Noah Kravitz. My guest is Nicolas Saricier. Nicolas is vice president of the three d experience platform r and d for TESOL Systems. We're here to talk about the next generation of Agentic AI systems, including industry world models, virtual companions, and the systems that are driving them. Nicolas, welcome to the NVIDIA AI podcast. Thank you so much for taking the time to join us. Thank you, Noah, and thank you for for the invitation and this opportunity to to be part of this, of this podcast. Absolutely. The pleasure is ours. So maybe we can start with you telling the audience a little bit about Dassault Systemes. They have a long running partnership with NVIDIA. So you can speak to that a little and then also to what your role is, and what the three d experience platform is. Okay. So so I'm Nicolas Soizier. I joined Dassault Systemes in 2004, and I'm now the vice president of, Swedish Smart Platform, Research and Development. And you have to know that the Swedish Smart Platform is really the foundation for our 12 brands, at Dassault Systemes. You know, things as the main brands, CATIA, SOLIDWORKS, SIMULIA, etcetera. And if you don't know us, we enable our customers to imagine, design, simulate, build almost everything in the world. Cars, airplane, autonomous robots, furnitures, electronic device, therapeutics, med devices, etcetera. It's 400,000 customers, 45,000,000 users, 15,000,000 scientists and engineers all around the world using our solution every day. And in fact, we provide our customers the factories to create their virtual twins. And what is virtual twins? It's really the scientific, multidisciplinary, multi scale representation of the product you want to deliver. And in fact, we enable a product to be tested in the virtual world, in the real condition before anything exist in the real world. And so today, my focus leading the Solid X Platform is really to transform our platform architecture into an agentic platform. And in fact, this is our shift from a SaaS platform, SaaS architecture to an agent as a service platform, to to bring AI to all our customers. So much has happened in in the world of AI in the past few years. And generative AI, obviously, has been, you know, this touch point that set off large language models and reasoning, and now we're talking about agentic systems. So let's talk about these two terms, virtual companions and industry world models. And what do those mean, to Dassault and the Dassault world? How do you use them? And how are they different from the types of generative AI that people might be used to using for the past few years? Yeah. So let's …

Get the full transcript (3,171 words) + summary by email — free

One-time email with the complete transcript and AI summary of this episode. No account needed.

One email, no spam. We’ll also show you what SignalCast does.

Browse all NVIDIA AI Podcast transcripts →

You just read a 3-minute summary of a 20-minute episode.

Get NVIDIA AI Podcast summarized like this every Monday — plus up to 2 more podcasts, free.

Pick Your Podcasts — Free

Keep Reading

Books, tools, and gear mentioned in this episode

SignalCast may earn commission on purchases via these links.

Tools

  • MarieBy guest

    by Dassault Systèmes

    Dassault deploys three specialized AI agents — Aura (business), Leo (engineering), Marie (science) — each reasoning through industry world models rather than general LLMs.
  • by Dassault Systèmes

    Dassault enforces full traceability of every AI interaction through a system called IP Lifecycle Management (IPLM), logging which workflows, models, and processes modified any content.
  • LeoBy guest

    by Dassault Systèmes

    Leo, for example, takes a 3D scan or 2D drawing, runs physics and kinematics analysis, and returns a manufacturable, optimized design.
  • by NVIDIA

    Nemotron reasoning models improved performance for Aura, Leo, and Marie by 20% without task-specific optimization.
  • by Mistral

    Dassault combines proprietary models with frontier models from NVIDIA (Nemotron via NIM) and Mistral, selecting partners based on performance and sovereignty constraints.
  • AuraBy guest

    by Dassault Systèmes

    Dassault deploys three specialized AI agents — Aura (business), Leo (engineering), Marie (science) — each reasoning through industry world models rather than general LLMs.
  • by NVIDIA

    Integrating NVIDIA NIM models improved document ingestion throughput by 30%, while Nemotron reasoning models improved performance for Aura, Leo, and Marie by 20% without task-specific optimization.
  • Open standards like MCP and agent-to-agent protocols enable cross-system agentic choreography without vendor lock-in.

More from NVIDIA AI Podcast

We summarize every new episode. Want them in your inbox?

Similar Episodes

Related episodes from other podcasts

Explore Related Topics

This podcast is featured in Best AI Podcasts (2026) — ranked and reviewed with AI summaries.

You're clearly into NVIDIA AI Podcast.

Every Monday, we deliver AI summaries of the latest episodes from NVIDIA AI Podcast and 192+ other podcasts. Free for one show.

Start My Monday Digest

No credit card · Unsubscribe anytime