Model-based False Annotation Detection: Developed Metric-based and Gradient-based optimization techniques for doing Loss Analysis to flag noisy labelled data. Improved the teams’ Data Quality by 3% and Transformer-based Table Structure Prediction model’s Test Accuracy by 1%.
Built an end-to-end Annotation Correction Tool to update mis-annotations with ML-assisted features.
Document Group Classification: Built an optimized and lightweight NLP model for classifying page into 250+ categories followed by page grouping module that completely discarded manual page-tagging task of users.
Developed low DPI document enhancement module, and multi-pass OCR for improvising character recognition.
Artificial Generation of Images: Worked on Unsupervised Deep Convolutional GAN and Conditional GAN for generating artificial OMR Sheets for increasing training set.
Image Classification: Automated the OMR-Sheet checking process by implementing image registration techniques followed by CNN having human-level accuracy.
Intern
Office of the Academic Registrar, Ahmedabad University
Duration: January – June 2018
Work done:
Documented the SDLC process of the in-house ERP system.
Reduce the new user on-boarding time by 40% via producing video materials.