Yogesh V Email & Phone Number

Proficient in developing Natural Language Processing (NLP) based Machine Learning models, as well as object detection models using TensorFlow/PyTorch tailored for Optical Character Recognition (OCR) from scanned documents. Independently, I engineered two Python-based OCR modules: one employing a fusion of hashing, deep learning, and other techniques to identify images and construct templates, while the other utilizes deep learning methodologies to extract text from distorted, skewed, and low-quality image receipts.This endeavor posed challenges where conventional methods like Tesseract and established deep learning models faltered. To surmount these hurdles, I conducted extensive experimentation, leveraging Graph Convolutional Neural Networks and vision encoder-decoder architectures. Utilizing transfer learning from widely accessible pretrained models whenever applicable. Additionally, I integrated appropriate OpenCV techniques into the process.Furthermore, I established a Flask API service for deployment on a remote production server, employing Gunicorn and Nginx. To streamline deployments across multiple servers, I devised Azure CI/CD pipelines connected to an Azure Git repository.In brief, I've encountered and worked with various tools and models for extracting key information from raw or scanned documents, including Python, MySQL, PySpark, Vaex, scikit-learn ML models, LIME for explainable AI, Keras OCR, CRAFT for text detection, transformer models, and hashing techniques.Although numerous OCR tools are available from tech giants, developing a tool entirely from open-source resources and striving to make it perform comparably is always exciting, pushing me to learn more. I have received recognition for innovative ideas that prove practical in product development. I remain open to learning, experimenting, and implementing open-source Language Model Libraries if the need arises.Additionally, I developed an analytics dashboard using Microsoft PowerBI, which entailed constructing intricate table relationships and implementing measures to a certain extent.