FINANCIAL DOCUMENT UNDERSTANDING PLATFORM
2021
Django
OpenCV
NumPy
BERT
TensorFlow
Hugging face
A platform to automate identification and information extraction from various financial and legal documents
• Developed a large-scale data pipeline for classification of 2M+ documents into 1400+ categories
• Worked on layout analyzer, table detection, table data extraction, forms and free text clustering
• Information extraction using BERT for Named Entity recognition in unstructured documents and ROI based template matching on structured documents
• Set up highly available, fault-tolerant, model pluggable infrastructure using AWS lambda, DynamoDB, API gateway
• Built an event-driven feedback loop and user interaction platform
• Reduced the SLA of single document processing by 70%
• Helped Intellect cut extraneous escrow staff and services cost by 34%
E-SIGNING & NOTARIZATION
2019
Signature, Notary Detection and Verification toolkit
• Designed and implemented a modular, extensible Signature and Notary Detection API
• Implemented a false positive removal api on top, increasing the model precision from 64% to 87%
• Developed an algorithm to validate if a notary seal follows US notarization state laws
• Setup data extraction pipeline from Notary Seals
• Integrated the api with DocuSign and First American document processing services platform
• Reduced the SLA of signature and notary verification by 70%
• Helped company cut escrow services cost, Increase in Document recording services revenue