At Tata Consultancy Services, I design and deploy document-processing pipelines for PDFs, scanned images, and unstructured data across BFSI, insurance, and legal clients. These deployments have reduced manual review effort by up to 50%.
I built document classification systems for more than 50 document types, achieving 70–90% accuracy depending on the type. I also trained and fine-tuned BERT models to classify documents by payer and document type.
I engineered reusable OCR pipelines with Tesseract, PaddleOCR, PDFPlumber, and Doctr. For document layouts and visual elements such as signatures and logos, I built CNN and DNN models using OpenCV and Faster R-CNN.
I customized spaCy and Flair NER pipelines to extract domain-specific information, and developed FastAPI and Flask REST APIs to serve models on internal servers. I also improved financial-document table extraction with Camelot.
