Work:

• Part of the OCR team at Karza, developed end to end training and evaluation pipelines for text recognition models and fine tuned text detection models for financial documents.
• Created data extraction pipelines from structered images pdfs using basic image processing for segmentation and simple heuristics for data validation.
• Worked on the deployment pipeline for the various developed deep learning models.

Work:

• Developed and fine tuned fashion image object detection models encompassing 20 classes spanning upper & low body clothes and footwear.
• Extracted the color of the detected object and mapped the color to its closest name.
• Deployed the models as REST APIs as microservices in docker containers.

Work:

• Developed a synthetic data generation pipeline for personally identifiable documents with focus to social security cards.
• Finetuned an object detector to locate and classify the card in an image.
• Redacted the PII data in the image by locating the text boxes and with simple heuristics to determine if box contains PII data or not.