I have built a Plug/Play system that can be integrated to turn any car into a Level-3 Autonomous vehicle. I presented this in an hackathon conducted by Mercedes Benz R&D, and won 1'st and got a chance to present the model in Barcelona, Spain for the MWC. Now, what we've done was build a Deep Learning Model that was trained on a car from the view of a driver , we've trained the model by driving the car inside a simulation for hours and once we reached required accuracy, we transferred the neural-network to a raspberry pi, now thi pi was acting as a brain of a rover which we've built. And the rover started behaving in the same way the the A.I was trained in simulation.
An another cool thing I've worked on is a device which can turn you into Magneto ! Just kidding, I've built a model for an EEG device which can classify your thoughts and perform respective action. The classifier was based on Deep Learning Model which can classify Brain Waves. My Deep Learning Model was trained on the EEG data that was collected using a device which is capable of absorbing brain's electronic impulses through the scalp of a person. And later when this trained model was introduced to real-time eeg data it started classifying thoughts like "Push", "Pull", etc.
I had to learn a lot of concepts regarding Deep Learning models which are based on "Long Short Term Memory" and dealing with a sequential data made the problem more complex for us than to dealing with regular and discrete data stream.