Projects
Context-based Sarcasm Detection
Arun Gaonkar, Manasi Ghosalkar
Learnt to use transformer-based RoBERTa and Bi-LSTM for text processing on a news dataset. (PDF, CODE, PPT)- Curated a news dataset of 28,000+ entries using BeautifulSoup. Pre-processed using pandas and numpy (ETL).
- Investigated context dependency for sarcasm detection by employing embedding & tokenization by NLTK libraries.
- Developed and trained Bi-LSTM & RoBERTa models for sarcasm detection, achieved 96% classification accuracy.
Example Bot – Personalized Code Assistant
Arun Gaonkar, Mohan Kanaka, Vinay Vasudev, Adarsh Narayanaswamy
A chatbot built by applying software principles using MongoDB, Express, NodeJS (MEN) stack, and Javascript. (Video)- Designed a chatbot to assist developers to create, store and retrieve personalized code snippets and API examples.
- Deployed server-based chatbot for CRUD operations on MongoDB by leveraging Ansible, Git bash & CI/CD pipeline.
- Improved test coverage to 96% by utilizing unit testing with Chai & Mocha following scrum & agile methodologies.
Wildfire Data Analysis and Cause Prediction
Arun Gaonkar, Ganesh Thanu, Nikhil Patil
Implemented ETL by scripting with pandas and numpy, learned data visualization techniques with matplotlib and seaborn. (PDF, CODE, PPT)- Led a team of 3 to build an end-to-end machine learning solution for wildfire cause prediction.
- Analyzed 1.88 million records using pipelined ETL, data visualization techniques such as matplotlib and seaborn.
- Employed models like RFC, KNN, Bi-LSTM, CNN to predict wildfire reason. Best accuracy of 93% was achieved by CNN.
Brain Tumor Image Classification
Learnt image classification and segmentation techniques by applying CNN, and R-CNN. (Video, PDF)- Developed MRI image classification models by training deep learning models like Bi-LSTM & CNN using TensorFlow.
- Improved classification accuracy to 92% by optimizing hyperparameters.