Abhishek Iyer

Abhishek Iyer

Masters in Robotics Systems Development at CMU | RA@IISc | Ex SDE@Design.AI | Interested in everything Robotics and Machine Learning.

Portfolio

Design.AI - Ensuring brand consistency

While at Design.AI, I was one of the core developers on making a multi platform AI backed tool to assist businesses in digitizing their brand guidelines and providing all designers instant evaluation on their designs with respect to the company’s brand guidelines. We also digitized Google’s Material Design 3 design system used by most freelancers. The plugin also offered suggestions for fixing violated guidelines, in addition to evaluating and fixing a few of the WCAG accessibility guidelines. Additionally, we provided a custom dashboard for managers to track team progress, improvement in efficiency, and business expenses saved. We further partnered with leading companies such as H&M and Kone to gain insights from their design teams on the Beta version of the tool.

Figma Creative Assistant

During my time at Aalto University, one of my main tasks was researching and developing a tool to assist UI designers explore color harmony. We used deep learning saliency models to extract colors from the focal points to build a palette around, which was extended using Monte Carlo Tree Search. The tool employed multiple assignment algorithms and heuristics in order to ensure light-dark variations, minimum contrast levels for accessibility, and a thorough user study. The tool was created as a plugin on the popular designing platform - Figma.

Marine Search and Rescue

Developing our own object detection models to aid in detecting, counting, and tracking people at sea during search and rescue missions. Further differentiation between floaters and rescue members can be made. Can extend this pronlem to tackle multi object tracking so even temporal data can be analyzed.

Lane Detection

Attempting to solve the issue of correctly identifying your current driving lane with a high level of confidence. The approach is to segment the current driving lane by using both deep learning and classical approaches and weigh the limitations of each method in order to assist with automated driving. Training and test data includes varieties of daytime, road curvatures, types of road markings, and surronding environment for robust learning