3 ms·
Startup Idea validation on ML end-to-end application deployment
I have a startup idea on providing a machine learning infrastructure for ML application deployment on distributed low-powered embedded devices such as Qualcomm, Nvidia, Xilinx, Google coral etc. With this ML users can deploy their Pytorch/Tensorflow model based multiple applications (python code) without writing any platform specific C++ code on any hardware device. This platform will also provide several tools to support ML life-cycle development to its production. I am curious to know if anybody will be interested in this kind of platform and services? Or should I think of providing custom services instead of a ubiquitous solution? Any feedback would be greatly appreciated.
- p1esk 5y agoCan you describe an example of your target user/application?
- akshack 5y agoTarget use-case: Deployment of several applications (such as object detection, semantic segmentation and body part tracking) on a single chip or on a distributed system. For example interior sensing and exterior sensing of a car, drones, manufacturing and assembling robots as well as service robots. Product: A lightweight multiprocessing and hardware abstraction framework for the deployment of multiple applications on a resource constrained embedded device which provides a common user friendly APIs and services across different devices.
- p1esk 5y agoDoes not sound very interesting tbh (I work in this space - developing power efficient DL accelerators). You should work on your pitch if you decide to pursue this idea.
- bisquequick 5y agoSounds like AWS Sagemaker no? Except this isn't cloud based and they also do optimizations on the model itself via aot compilation.