Information Extraction from unstructured documents: Implemented a generalised one-shot Learning model for extracting 80+ Singular and Complex Tabular fields from documents that reduced manual data-entry work from 20 to 5 minutes per document set.
Zero-shot Learning: Observed the annotation challenges faced by the users and proposed & lead information retrieval on unseen documents using NLP and CV that increased job annotation speed by 3X.
Self-learning based feedback mechanism: Built and deployed Federated Learning and Reinforcement Learning inspired model to improvise data extraction and standardisation using online-learning from data collected from all the clients respecting data privacy.
Other Responsibilities: Mentor junior developers, sprint planning, feature prioritisation, client handling, conducting UAT, interviewing for mid to senior level roles