•Developed a short answer grading system using tree-based classification algorithms to achieve 89% accuracy on unseen data. Engineered features to capture semantic similarities between question-answer pairs.
•Devised a custom page-number recognition system employing the contouring function in OpenCV and a CNN with less than 100k parameters to achieve 99% accuracy on unseen data using transfer learning
•Delivered software APIs for AutoGrader, a platform for automated student assessment and feedback, in Python, adhering to OOP principles. Wrote unit tests and documentation for production-level code.
•Improved performance of an inference pipeline for a Resnet model in production. Reduced runtime of a bulk pdf-processing module by ~50% using multiprocessing in Python.
•Built a content-based filtering system for an e-commerce website to alleviate the cold-start problem. Generated top-N candidates using cosine similarities between product descriptions scraped from the client website.
