The Edge AI Foundation is pleased to welcome Kernwerk as our newest Startup Member!
Fresh off its win at the Edge AI Taipei 2026 Startup Pitch Competition, Kernwerk is bringing a practical approach to one of the industry’s most persistent challenges: getting AI models to run efficiently and securely on the right hardware.
Based in Berlin, Germany, Kernwerk is developing a platform that helps AI developers and device manufacturers test models on real hardware, optimize them for specific processors, and protect their intellectual property once deployed.
Its mission can be summarized in three words: Test. Optimize. Secure.
“We’re excited to be part of the community, learning from the people who put AI on real devices every day. It’s an honour for us to win the EDGE AI Taipei pitch competition, especially as the typhoon was slowly landing over our heads! Our vision is simple: the best device for the job, an optimised model, safely deployed. And we’re looking forward to working on this with the community.” – Pierre Dubouilh，CEO & Co-Founder
Bridging the Gap Between AI Models and Real Hardware
As edge AI adoption accelerates across robotics, medical devices, computer vision, automotive systems, and industrial applications, developers face an increasingly complex hardware landscape.
With more than 20 vendors offering edge neural processing units (NPUs), selecting the right chip is no longer as simple as comparing specifications.
A processor’s advertised TOPS (trillions of operations per second) may indicate theoretical computing capacity, but it doesn’t necessarily reveal how a particular AI model will perform in real-world conditions.
Developers must navigate multiple trade-offs, including inference latency, energy consumption, thermal performance, memory usage, and model accuracy.
Meanwhile, deploying proprietary AI models on distributed devices introduces additional security concerns, from model extraction to unauthorized replacement.
Kernwerk addresses these challenges through a unified, hardware-aware workflow that connects model evaluation, optimization, and security.
From Model to Deployment: Test, Optimize, Secure
Kernwerk’s platform focuses on three essential stages of edge AI deployment.
1. Test on Real Hardware
Rather than relying solely on datasheets or theoretical benchmarks, Kernwerk evaluates AI models on physical development boards.
This allows developers to compare candidate hardware platforms using measurements relevant to their actual models, including latency, power consumption, accuracy, and deployment considerations.
The goal is to help engineering teams identify suitable hardware before committing to a particular processor or architecture.
2. Optimize for the Right Chip
Once a hardware target is selected, Kernwerk focuses on model optimization.
Its approach incorporates techniques such as INT8 quantization, pruning, memory optimization, scheduling, and hardware-aware configuration.
These optimizations help developers navigate the trade-offs between inference performance, resource utilization, and model accuracy.
Instead of treating AI model optimization as a one-size-fits-all process, Kernwerk focuses on the capabilities and constraints of the target hardware.
3. Secure AI Models on Devices
As edge AI systems move from prototypes into commercial products, protecting deployed models becomes increasingly important.
Kernwerk is developing deployment security capabilities designed to reduce the risk of proprietary model extraction and unauthorized modification.
By integrating security into its deployment workflow, the company aims to help device manufacturers protect valuable AI intellectual property throughout the product lifecycle.
Building a Hardware-Aware Intelligence Layer
Kernwerk’s work extends beyond individual model deployments.
Its platform analyzes how neural network operators execute across different hardware architectures, identifying operations that can run on NPUs and those that fall back to CPUs.
These insights are valuable not only for AI application developers but also for semiconductor companies designing the next generation of AI accelerators.
Kernwerk’s longer-term vision is to use operator-level performance data to inform future NPU architectures, validate computational primitives through FPGA experimentation, and explore increasingly specialized hardware implementations.
This creates an opportunity to connect two sides of the edge AI ecosystem: the developers building AI applications and the semiconductor engineers designing the hardware that powers them.
From Edge AI Taipei Winner to Foundation Member
Kernwerk recently took home the top prize at the Edge AI Taipei 2026 Startup Pitch Competition, demonstrating the relevance of its approach to the broader edge AI community.
Its participation reflects a growing industry emphasis on the practical realities of edge AI deployment: hardware compatibility, measurable performance, optimization, and security.
As a Startup Member of the Edge AI Foundation, Kernwerk joins a global network of startups, semiconductor companies, technology providers, researchers, and industry leaders working to advance edge AI adoption.
We look forward to connecting the Kernwerk team with hardware manufacturers, device makers, ecosystem partners, and investors who share their vision of making edge AI deployment more accessible and reliable.
Meet the Team
Kernwerk brings together expertise in embedded systems, security, and machine learning.
Pierre Dubouilh — Embedded Systems & Security
Pierre brings ten years of experience in low-level systems and security, including work at Fastly and Thistle.
Maxime Carriere — Artificial Intelligence & Machine Learning
Maxime holds a PhD in AI and brings ten years of experience working with machine learning under tight computational constraints.
Together, they are building tools to make the transition from AI model development to real-world hardware deployment more systematic and efficient.
Building the Future of Edge AI, Together
Kernwerk is actively seeking opportunities to collaborate with device manufacturers looking to test, optimize, or secure AI models, as well as module manufacturers and ODMs interested in expanding hardware testing capabilities.
The company is also engaging with investors as it works to scale its hardware lab and engineering team.
At the foundation, we believe that the future of edge AI depends not only on more powerful models and processors, but also on the tools that make those technologies work together.
Welcome to the Edge AI Foundation, Kernwerk!