Jul 2022 – Dec 2022 · 5 mos · Pune, Maharashtra, India

Aug 2021 – May 2022 · 9 mos · Pune, Maharashtra, India

• Devised a classification model for auto-matching relevant legal documents, along with a web app for seamless user interaction and document retrieval. Achieved a 20% increase in recall rate by training on a continuous data stream
• The app reduces the manual work of document classification and initial training set size of traditional machine learning models
• Reduced initial training data requirement by 65% by building an ETL pipeline on Spark to leverage training on incoming stream of data
• Solved the problem of catastrophic forgetting using pseudo-rehearsal faced during incremental learning by training over incoming data stream

Jun 2021 – Sep 2021 · 3 mos

• Automated manual classification of bug reports by developing a deep learning-based model based on BERT and Bi-LSTM
• The project automates the manual bug classifications, earlier done by customer service reps of UST
• Achieved classification accuracy of 80% over 3 bug classes and 76% over 4 classes
• Developed a deep learning-based model to classify bug reports. Extracted sentences containing keywords from raw incomprehensible dataset to feed into neural net.

Mar 2020 – Jun 2020 · 3 mos · Raurkela, Odisha, India

• Created a new twitter classification model to attain up to 10% increase in accuracy over previous state-of-the art models
• Fine-tuned BERT with Bi-LSTM on top, combined with a global max pooling layer

Aug 2019 – Oct 2019 · 2 mos

• Developed and deployed main admin application to control user-side application. Unit tested admin app and user-side app
• Increased conversion rate by 15% by integrating UPI payments & Razorpay in user-side application

Oct 2018 – Dec 2018 · 2 mos · Pune, Maharashtra, India

• Collected and managed the relevant data to be published on the website.
• Converted mathematical equations into computer readable format using Latex language.