My portfolio showcases various projects created throughout my career. Most of these projects were done in my undergrad to participate in national and international competitions. These projects also cover some research work done as a part of my PhD course work final projects. See my contact information below and get in touch.
BongoBeer – Your Personal Assistant AI Robot
Meet BongoBeer, a robot you can speak to, communicate with and post any question in English or Bengali. It is a robot accompaniment fitted with sophisticated Artificial Intelligence. Also, it will handle different tasks for you as seen in the video demonstration.
Project Vision: Smart home security system using capacitive sensing
This system introduces a 3D capacitive sensing system that will make an entire room a capacitive sensor after applying capacitive paint in the room (floor, roof, and one wall). As we know, human body is capacitive, and the capacitance of a parallel plate capacitor is proportional to the area, A in metres2 of the smallest of the two plates and inversely proportional to the distance or separation, d (i.e. the dielectric thickness) given in metres between these two conductive plates. When people enters the room and walks, the distance from human body and all three painted walls (acting as capacitor plate) changes. Thus we can observe a change in capacitance and detect the position of the person in the room. There is also a camera in the room, which will in general be turned off, and only active when there is a human presence. It can detect face and match with previously reported family members faces. When there is a human presence from capacitive sensor and the face does not matches, it sends a notification message to the owner with the image and video of the person roaming in the room. As the capacitive sensor can detect the human position in 3D the camera can move to detect the face of the intruder properly.
Energy Efficient Lifestyle at Household Level
Power crisis and shortage of electricity is a big problem in developing countries like Bangladesh. So, we came up with following solutions as shown in the video.
Reinforcement learning has made great strides in exploring virtual environments, especially those of video games because of their well-structured reward systems. Training time is a crucial aspect of reinforcement learning since it influences factors such as adjusting for parameters, prototyping, and eventual deployment. However, the training times required for an agent tend to be very high, mainly due to sparse rewards received from the environment. Agents operating in complex environments have further computationally expensive tasks such as the exploration of high dimensional state spaces. We propose replacing sparse environmental rewards with human-defined rewards, that leverage the wisdom of crowds to accelerate the training of machine learning agents. We reduce the training time further by using a multi-agent setup along with their corresponding human trainers that would allow knowledge to be pooled from agents with different experiences. We define new mediation strategies and use these along with previously known strategies for such voting.