Tools : Python - PyTorch, Keras, Flask, Tensorflow, SQLite, CSS, Bootstrap

• Built an application to extract transcripts from video lectures and provide short summaries and deployed it as web application with Flask
• Used a Transformer model with pretrained BERT tokenizer to summarize text transcripts, PyTorch Silero models to extract text from audio files and ffmpeg to extract audio content from videos
• Extended the web application to serve as a platform for professors/teachers to upload videos for students with automatic summarization to aid with remote education

Tools : Python - PyTorch, Keras, Flask, Google Cloud Platform - Cloud Run, Cloud Build, JavaScript, CSS, Bootstrap

• Built a BERT based model with PyTorch and Keras preprocessing libraries to detect the probability that a given news article is fake.
• Deployed the model to an application server built with Flask to serve requests from frontend applications. Also deployed the application server to Google Cloud Run by setting up automatic CI/CD pipelines from Github repository.
• Created a Chrome Extension to serve as a frontend with JavaScript and Bootstrap to allow users to verify the presence of fake news in articles online.
• Used available APIs to suggest alternative sources of information when articles with high probabilities of being fake get accessed.

Tools : JavaScript - React, d3.js

• Analyzed the details of space missions launched worldwide since 1957.
• Designed and developed interactive visualizations implementing techniques such as highlighting, brushing and filtering using d3.js and React to showcase the analysis.

Tools : Python - PyTorch, OpenCV, Matplotlib, Pandas

• Detected melanoma among images of benign and malignant skin lesions from the SIIM ISIC Melanoma Challenge Dataset.
• Used EfficientNet for feature extraction, incorporated metadata, performed data augmentation, and image preprocessing to crop out regions of interest from the images. Achieved 80% accuracy.

Tools : Python - PyTorch, OpenCV, Matplotlib, Pandas

• Performed multi-class classification to identify all the types of proteins present in the cell images from Human Protein Classification dataset.
• Performed data pre-processing and data augmentation, used Transfer Learning with pre-trained ResNet50 model to make predictions.

Tools : Python - NLTK, Pandas, Numpy

• Created a search engine to fetch the most relevant reviews to the given user query using BM25.
• Performed topic modelling using LDA.
• Generated more accurate user ratings based on Vader sentiment analysis of the user reviews.

Tools : Python - NLTK, Pandas, Numpy

• Performed data mining of Twitter data using twitter API and tweepy library for tweets containing the keywords “Donald Trump” and “Joe Biden”.
• Performed exploratory and sentiment analysis to figure out the general sentiment for each candidate and predict the winner of the 2020 presidential elections.

Tools : Android Studio, Java, SQLite

• Developed an Entity Relationship Model and a Database using SQLite to capture the customer and travel details such as preferred transport, hotel, restaurant and tourist attraction.
• Created an Android application to allow user to plan the trip and view their itinerary.

Tools : Python - NLTK, BeautifulSoup

• Crawled webpages using BeautifulSoup, performed text preprocessing using natural language processing (NLP) techniques.
• Implemented PageRank algorithm to rank the webpages.

Tools : Scala, Apache Spark

• Used Apache Spark and Scala to optimize the join operation between two large data sets (13M and 0.1M records) with one of them skewed.

Tools : Python - Numpy

• Performed feature extraction and developed a human physical activity recognition model for predicting the activity performed based on the person’s movement data.
• Evaluated and compared several statistical learning methods including Logistic Regression, Random Forest and Support Vector Machine.