Selected Projects in Machine Learning Projcts

This project is about automating the detection of a text from its contents if its safe or harmful in any sence.
This is a multi-label problem with a huge imbalance in classes.
We use Distill-Bert Model which is a distilled form of BERT made by knowledge distillation.
For training we use mixed precision training where precision is set to 16 bits.
To furthur reduce the model size we use optimizatiion methods such as quantization where the model weights are converted from 32 bit float to 8 bit integer.
The evaluation method used here is AUC-ROC score per label.
To make the model language agnostic, I use ONNX.
Currently to deploy it a Flask server is used.

This project is about extrating data from financial documents and classifying the type of document it is.
This is a object detection, recognation and imbalanced image classification problem.
For object detection and recognation currently easyocr is being used. For image classification layoutLMV3 is used.
For training we use mixed precision training where precision is set to 16 bits.
Used docker for training and deployment env seperation.
Deployed it using streamlit.

Experience

Associate Software Engineer at HighRadius

Software Engineer Intern