Student Resource Management

A java GUI application used for economically and swiftly sharing books, notes and equipment amongst students.
The project followed MVC design pattern and Waterfall SDLC model.

Keyword Spotting System

A 10 keyword class audio dataset was preprocessed into MFCCs. And a CNN model was trained on that data and integrated into a Flask app.

Analysis of Manhattan neighborhoods for business feasibility using
Foursquare API

It’s a capstone project of ‘IBM Professional Data Science Track’. It involves web scraping, data cleaning, feature engineering, data querying from FourSquare API, Map visualization using Folium library and using K-Means clustering to analyze the neighborhoods/clusters.

Image Colorization using Deep Convolutional Neural Network

A autoencoder network to colorize black and white images, trained on a subset of Places365 dataset. The model used Vgg16 CNN network as encoder and the extracted features are passed to the decoder.

Santander Price Prediction

Developed a neural network model for the price prediction task in this Kaggle competition.
The model scored a AUC-ROC score of 0.92027 on the testing data.

Movie Recommendation System(webapp)

The recommendation system uses content based filtering on the ‘TMDB 5000 Movie’ dataset in python. Count vectorization and Cosine similarity are used for recommending top 5 similar movies. And sreamlit library is used to design the webapp and TMDB API is used to query movie details. The webapp was also deployed on Heroku cloud.

Neural Style Transfer

Implemented the deep neural network(Generative Adverserial Network) from scratch

Automatic text generation using Deep Learning

It is a deep learning model using LSTMs and Neural Network to generate sequence of text after being trained on a corpus of Shakespeare’s work.

Deep Learning Models

Implemented deep learning models from scratch from the papers.

  1. Inception Net
  2. ResNet
  3. LeNet
  4. Vgg16