Projects
​
Projects have been an integral part of my learning journey. Whether it be learning how to use Scikit-learn properly or understanding a concept, all of the aspects have revolved around projects and POC's. Until now, I have completed a lot of projects which have helped me understand myself and my areas of interest. Some of my research papers started as small homegrown projects which later took shape as publications. I have cherry-picked three projects to showcase here. ​ ​ ​
K-Fold-Imblearn
KFoldImblearn is an open-source python package that handles the resampling of data in a k fold fashion, taking care of information leakage so that our results are not overly optimistic. It is built over the Imblearn package and is compatible with all the oversampling as well as under-sampling methods provided in the Imblearn package. It is published on PyPI and is currently being used by a few data scientists and machine learning practitioners.
​
Link to Documentation: Click here.
Link to Code: Github
Jugni
Jugni is my very own virtual assistant. It is a python-based end-to-end virtual assistant. There are numerous repetitive tasks in our day-to-day lives which VA's like Siri and Alexa cannot handle. Jugni is a fully customizable VA in which you can plug and play python modules to get your tasks performed. A link to the POC has been provided below.
Link to Code: Github
Demonstration Video: Youtube
Automated Music Composer
Since I myself am a musician, I exactly know the challenges and blockers that we musicians face while writing and composing new songs. The motto behind this project is to minimize the ideation time of a musician/producer to be negligible. Songs we hear in our day-to-day lives are nothing but musical ideas which are polished and well prepared as we hear the final product. What if we provide music producers with plenty of ideas with just a single click?
​
This POC consisted of 4 RNN based models (one model for each element: melody, chords, beats and percussion). The data was self-prepared using melodies and chords from artists like Ed Sheeran, Coldplay, Alan Walker and more. The training data was in a MIDI file format (sequence of notes) and the output produced is a WAV signal of the output.
​
