With DavinSy, the transfer of learning to the terminal becomes possible. This opens a whole new range of professional applications and helps boost the cost effectiveness and performance of today’s low power electronic systems.


iDetect-4.0 is dedicated to helping industries improve their production quality, diminish their Cost of Poor Quality and prevent production shutdowns.

Limitations of current AI-based quality control solutions

  • Current solutions suffer from their inability to adapt to the constantly changing conditions encountered in factory plants. Based on iterative backpropagation learning algorithms, the lifecycle of the models is too slow to keep pace with the rhythm of environmental changes. 
  • Similarly, successful proof of concepts on one machine are often not replicated when deploying across the entire inventory.
  • Furthermore, model maintenance often relies on heavy cloud infrastructure and expensive, and rare hardware accelerators. The necessity of model maintenance is frequently realized too late, and the accompanying hidden costs weigh heavily on return on investment, making such solutions unsustainable.
  • Finally, manufacturers, become very dependent from external expertise and resources. This weakens them, putting them at the mercy of decisions they do not control.

Benefits of iDetect-4.0

  • iDetect-4.0 relies on DavinSy agents. Thus it is very agile and can adapt, with the active help from the operators, in real time to any environmental change.
  • This same capability enables the deployment of successful PoCs on other machines. After a brief learning period, the new instance will be as efficient as the original one.
  • Everything is done locally on a standard PC (no need for a GPU) and can be managed through a simple interface by the quality team or experienced operators.
  • Manufacturers retain control of the software. They can configure it autonomously through Maestro. They can generate as many instances of iDetect-4.0 as needed.



DavinSy Voice is dedicated to markets such as autonomous robots or cobots controlled by voice commands, voice-activated hearing aids for medical and consumer applications, voice control for minimally invasive surgery, identity theft prevention with biometrics speech, cough analysis to identify pathologies, smart home voice control, as well as predictive maintenance of airborne noises.



DavinSy has been evaluated and can be configured for precise motion control systems based on inertial motion sensors. These markets include the extended precision radar system for self-driving cars, drone navigation systems, injury prevention by analyzing the biomechanics of the runner’s stride, real-time monitoring of the engine power of the electric car.

Limitations of current AI-based motion control solutions

  • Poor real-world performances due to the complexity of building an exhaustive training dataset
  • Model drift due to high variability of the captured real live data
  • Increasing data volume and complex network optimization methodologies requiring cloud-based high performance computing solutions
  • Generic complex models imposing a vicious endless loop between overfitting, more data to avoid overfitting, addition of new network parameters to account for the new data variabilities, and the new parameters needing optimization
  • Reduced autonomy due to dependence on the cloud (white area)

Benefits of DavinSy

  • Unique continuous learning feature from environmental variations for advanced failure anticipation
  • Unique Virtual Model concept avoiding the problem of static models drift over time
  • Unique hyper-miniaturized footprint extending runtime for battery-powered applications
  • Reinforcement learning increasing the intelligence of voice control systems
  • Better loyalty and confidentiality of data guaranteed by the locality of the treatments
  • AI engine managing multi-model and multi-modal architectures allowing intelligence to be distributed over several connected devices
  • Minimal latency on both learning and inference thanks to full cloud independence



DavinSy with its communication features strengthen the security by a better collaboration between DavinSy AI engines solving multi-modal/multi-model problems. This increases the local security by distributing AI over multiple existing connected devices such domestics robots, access controls combining face recognition, voice biometrics, and potentially footprint keys within smart home applications. All privacy sensitive and critical data industrial applications will benefit the local and secured database management of DavinSy.

Limitations of current AI-based operations solutions

  • Poor real-world performance due to data variabilities
  • Poor rejection of unknown classes (impostors)
  • Cloud dependencies and related privacy issues due to necessary data mirroring between cloud and edge

Benefits of DavinSy

  • Unique mathematical algorithm immediately blocking intruders or unusual events
  • Unique virtual model concept adapting real-time security level to unexpected behaviour
  • Hyper-miniaturized footprint of the algorithm making it possible to distribute DavinSy over all terminals and reinforcing the least resistant access point
  • Total cloud-free autonomy allowing for rapid training, model adaptation and increased safety



DavinSy with its data-driven continuous learning capability is able to handle a large range of environment variability and still adapts its virtual model just-in-time to maintain the accuracy of the systems. This unique characteristic makes possible to calibrate sensors taking care about potential deviations over the time by continuous monitoring and model adaptation to the changes in the chain of the components. This serves markets such as failure detection and prevention in predictive maintenance, injury prevention in healthcare, as well as precision farming applications.

Limitations of the current AI based Voice-control solutions

  • Poor performances in real life conditions due to complexity to learn from the real live data
  • Data augmentation and network optimization methodologies leading to necessary high computing power unsuitable with battery operated systems
  • Raw database and models size exceeding the available resources of connected devices

Benefits of DavinSy

  • Learning from nominal usage permitting the imputation of missing measurements for failling sensors 
  • Unique virtual models concept extending the durability of the model to product life
  • Unique hyper-miniaturization footprint securing longer autonomy on battery operated applications
  • Unique reinforcement learning feature accounting for the specificities of the chain of components
  • Autonomous processing and local database for increased safety and accuracy in failure detection 
  • Real-time learning and inferences thanks to DavinSy cloudless algorithm

Meet Bondzai