1000 Designs a Day: Neural Concept's Thomas von Tschammer on AI-Native Engineering
Summary
Neural Concept, a Swiss company, utilizes specialist AI models for engineering domains such as aerodynamics, heat dissipation, and collision safety. Thomas von Tschammer, co-founder and US Managing Director, highlights how these models accelerate product design for automotive manufacturers like Jaguar Land Rover. Historically, design progressed from manual drafting to physics-based digital simulations, which, while faster, created compute bottlenecks. Neural Concept's models now provide simulation results in minutes, allowing manufacturers to evaluate over 1,000 designs daily, a substantial increase from 50. Their Engineering Copilot integrates agentic optimization with domain-specific validation, enabling direct CAD platform interaction for design changes. This frees human engineers for complex trade-offs and occasionally produces "Move-37-like" designs surpassing human intuition. The company aims to develop general-purpose foundation models for engineering, extending beyond per-customer training, with applications even in Formula 1 racing, where compute for aerodynamic optimization is capped. This AI-driven evolution promises faster product cycles and enhanced efficiency.
Key takeaway
For Directors of AI/ML or VPs of Engineering aiming to accelerate product development, integrating AI-native engineering workflows is critical. You should prioritize implementing AI models for core disciplines like crash safety and aerodynamics to achieve 20-40% speed-ups in year one. Subsequently, focus on cross-disciplinary orchestration to break silos and compound benefits, potentially reducing overall development cycles by 50-60%. This transformation is essential to remain competitive against agile, AI-first companies and avoid exponential gaps in product cycle times.
Key insights
AI-native engineering, employing specialist models and agentic workflows, dramatically accelerates product design and uncovers novel solutions.
Principles
- AI models learn intuitive physics from simulation/test data.
- Agentic workflows combine optimization with validation.
- AI-driven design enables exploration beyond human intuition.
Method
Neural Concept's Engineering Copilot uses domain-specific prediction models as tools, interacting with CAD platforms to make design changes. This agentic optimization and validation loop is ideal for Reinforcement Learning.
In practice
- Conduct 1,000+ aerodynamic tests daily.
- Reduce battery cool plate development cycles by 80%.
- Find designs 2-5% more aerodynamic.
Topics
- AI-Native Engineering
- Product Design Acceleration
- Computational Fluid Dynamics
- CAD Integration
- Reinforcement Learning
- Automotive Engineering
- Formula 1 Aerodynamics
Best for: Executive, AI Product Manager, Product Manager, Director of AI/ML, VP of Engineering/Data, Consultant
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by The Cognitive Revolution.