RAUSHAN KUMAR

Designed, developed and optimized ML and DL solutions in area of NLP, OCR, text classification, information extraction, image processing. Designed and deployed Duplicate Document Identification solution for an Australian bank. Managed application development life cycle, requirement gathering, technical design, coding and testing. Developed models using Hashing, convolutional autoencoder and ANN with a client server model deployment pattern. Converted Machine learning models into services and pipelines to be consumed at web scale on AWS cloud. Optimized application code for GPU utilization to achieve maximum speed and scalability. Recruited and trained 10+ ML engineers and supervised the projects where they were aligned. Implemented a pdf digitizer (OCR, restructuring of text, color identification and QR code identification).

Developed a deep learning solution to extract information from semi structured / structured documents (eg. Invoice and Cheque processing). Developed model using Bi-LSTM network to learn neighborhood information. Implemented Word embedding (word2vec & Fast text) to avoid OCR error in text conversion. Implemented an Auto tagger to create training data for LSTM network in order to reduce manual effort in NLP tagging. Developed a solution to extract tabular data from tables like structure (Balance sheet of financial documents/ Line items of invoices). Build an Image segmentation model for text box detection in images using EAST model. Build and trained Multilingual OCR for document processing. Implemented a space normalization and regression based restructuring algorithm to retain structure of images in text file. Implemented orientation/skew identification and correction module for images. Implemented logo identification in images to classify documents.

Developed a text based classifier for classifying medical prescriptions using TF-IDF followed by Neural network and cosine similarity to identify prescribed devices. Implemented an unsupervised clustering technique when no of classes is not known (Incremental K-means). Implemented an unsupervised clustering technique for clustering image documents. Implemented parallel programming for the existing code base and convert them into Object oriented code base.

Designed, developed and automated complex application to extract, transform and load data in order to meet customer requirements.