Transformers are a novel neural network architecture that have grown in popularity to solve problems such as Natural Language Processing. There is a need for hardware capable of running inference on transformers with higher throughput and energy efficiency to maximize their effectiveness. Socotra is a hardware and software stack for running Transformer neural networks on Field Programmable Gate Arrays (FPGAs). FPGAs offer flexibility in hardware design to perform inference with superior power efficiency at a competitive throughput.

Maintaining consistently cold temperatures during vaccine transport & storage is vital to maintaining vaccine effectiveness. Countries with weak public infrastructure often have inadequate cold storage capability due to poor equipment and lack of expertise. Domestic fridges are unusable because their poor temperature regulation leads to vaccine losses. Retrofrigeration is a fully integrated kit that retrofits existing domestic fridges with improved temperature control systems to bring them up to WHO vaccine storage standards, allowing regions with a lack of resources to reduce their vaccine wastage and improve public health.

Engineous is a mobile application that provides driving quality and road infrastructure analysis. Driving data such as acceleration and position are gathered using smartphone technology and sent for analysis on a cloud-driven backend. Statistical methods and machine learning models classify the data into important events, such as sharp turns and potholes, and score the driver on their performance. This reported back to the user to educate and improve driving performance. The collected data is also valuable to track road conditions for improvement operations.

Teletunes is a technology that provides an accessible way to control music playback interactively for those physically unable to do so on their own. Such people include those with debilitating spinal issues, poor motor control, amputees, etc. Teletunes is a headset paired with an application that interprets neural signals to control Spotify. It eliminates the necessity of manual control providing a hands-free solution. The project interfaces a 3rd party neural signal IC, and leverages concepts of machine learning, distributed computing, and integrated circuit design.

Humans require about 3-4L of water per day, but are often too busy to build good hydration habits. Our solution to this problem is HydroBuddy, an IoT connected smart lid attachment for water bottles. Once attached, HydroBuddy tracks the amount of water consumed by the user throughout the day, providing intuitive feedback and metrics through an IOS app. Additional features include temperature sensing and water quality sensing.