 So let's move on guys. Maybe the last little thing we also hear show on AI is similar to what you have seen with our MPU solution with the X Linux AI package but here we are running also people detection and people counting not on the microprocessor but on the microcontroller and I let Guillaume to maybe say a few words about the demo and the best advantages of this So here we have two example applications that we can run through our tool called STM32 QBI which allows you to convert pre-train neural networks into optimized C code for the STM32 microcontrollers. So the beauty here is that you would typically see those kinds of applications on MPU and or a doorbell and it will detect the faces and then for each face you can compare it against a database of enrolled people. So right now I am not recognized but if I go in front of the camera I can enroll my face and now I'm recognized as user too. Anybody else who will go in front of the camera will not be known by the the system. QBI is a software component and you can use pre-trained model from TensorFlow, Keras or PyTorps True ONNX and then generate the code and integrate it into your application. So the QBI is a very exciting field to specialize in. Yes and this is the beauty. Typically you would require MPU and now we can start to see some applications running on microcontrollers with very low memory and small uh how's the performance? The performance is we're starting to see some decent numbers so right now at four FPS a few years ago we were running only at one FPS so every every year we're making some progress. Is somebody making some magic in the background to make it faster? We have a very dedicated team that is optimizing our QBI library to take advantage of the DSP instructions and dual-issue pipeline of the Cortex M7 microcontrollers. And there's even more stuff that could be optimized? For sure. All those? Yes. Okay thank you.