Science of Reading made easy. Generative AI-powered personalized learning for early readers, educators and parents.

• Full stack development of DeepEngage, a tool that combines ResNet34 with a categorically-embedded tabular network to help companies maximize the reach of their sponsored posts in target niches by forecasting post engagement on Instagram• Developed a data collection and processing pipeline consisting of: data scraping >30,000 Instagram posts and their associated metrics using selenium, data cleaning using pandas, and feature engineering using imageio and numpy• Automated the visualization of niche performance validation using a combination of pandas, matplotlib and seaborn, revealing that DeepEngage is able to tag poorly-performing posts in the lifestyle niche with 55% precision (vs. 12% for random guessing)

• Created and optimized the first homoFRET-based biosensor, which reduced the number of colours required for ratiometric fluorescence detection from 2 to 1.• Built an image processing pipeline in python that used sequential deep learning models (YOLO andU-Net) to automate data extraction from biological images, reducing the human labour required by>98% and providing a platform to explore and summarize data using bokeh• Designed a custom dataloader for fastai to load subimages from mixed biological image stacks, allowing a direct comparison of segmentation approaches and reducing the required size of the training set by two orders of magnitude• Wrote and was awarded a $90,000 grant based on previous work to design and build a novel microscope for high-throughput image collection and analysis, increasing the number of cells collected per experiment from 10-30 to >10,000• Spearheaded the creation of an internal machine learning group to apply machine learning principles to biological image analysis.• Directed the training and project development of two undergraduate thesis students and three summer students

Courses Taught: Systems Engineering, Mol. Biophysics, Biomolecules and Cells

• Designed and constructed a digitally controlled signal generator for a HIFU system, increasing the usable channels from 1 to 16, as well as the software required to calibrate the device (Labview/C++).• Design and ran characterization software for piezoelectric transducer units.