Harish Soni

Engineered AI technology to solve complex Healthcare problems both supervised and unsupervised. Improved text-based insight extraction from large documents using T5 architecture. Researched Modality Classification resulting in 10% more useful insights from images into textual data. Conducted NLP-based (RNN) and Computer Vision-based (CNN) research projects with statistical analysis. Proficient in data processing and manipulation using Pandas, NumPy, and Scikit-learn and back-end.

Managed back-end software product life cycle including design, API implementation, and relational database management. Utilized SOTA Yolov5 and deep sort algorithm for vehicle detection and tracking. Developed applications using Python, PyQt5, PyTorch, and Computer Vision, OpenCV. Managed multiple model pipelines and using hyper parameter tuning for regression algorithms to achieve over 93% r2 score. Collected data from multiple sources using web scraping techniques NLTK. NLP Techniques cover: Lemmatization, Stop Word Analysis, Word Tokenization, ocr, and other.

Contributed to a Computer Vision based project that converts hand drawn image to an html file. Reduced overall data preparation time by 10-15% by assisting in generating synthetic dataset for model training. Developed a custom HTML generator and played a role in model building in Detectron2. Designed and implemented data pipelines using Data Science technologies to overall project architecture.