Hi! I'm Dhruuv

About Me

Love to keep learning new things and pick up skills and hobbies along the way.

Currently in the final semester of my Master's Degree in Data Science at Indiana University Bloomington. Passionate about Deep Learning, Machine Learning and Data Science. This drives me to venture into different projects and domains. Do check out my work on Github and Kaggle. Open to colloborations and suggestions!

Experience

Image Analytics Intern-NLP | Siemens Healthcare

Malvern, Pensylvania, USA
  • Improved efficiency of existing patient database retrieval by harnessing text information instead of processing images
  • Conducted research to come up with two novel ideas to infer patient body characteristics. The proposed method, built in Keras, utilized both character and word level information and used a combined loss of Cross Entropy and Dice
  • Presented the SOTA work, XLNet, to scientists at a Seminar and discussed its developments in Language Understanding

    • Research Assistant, Machine learning | Indiana University

    Bloomington, Indiana, USA
  • Explored plant gene data to build a model to identify elements in promoter regions, crucial for transcription

    • Systems Engineer | Siemens Healthcare

    Bangalore, India
  • Worked on TrueD, part of Syngo Classic platform, which allowed doctors to compare and assess diagnoses of a patient
  • Developed and maintained the application to render the 3D image reconstructed from diagnostic MRI and CT scans
  • Led a team of five in initial project to develop a chat console application using Named Pipes with C++ 11 features
  • Migrated the entire code base of the TrueD application written in C++, from 32 bit to 64 bit version

    • My Projects

    Kaggle-Dog Generation GAN

  • Secured

    global rank 31/927

    , with MFID score of 110, in task of

    generating realistic dog images

    for Stanford Dog Images dataset.
  • Tweaked Deep Convolutional Generative Adversarial Network with mapped dog breed as condition

    (cDCGAN)

    • Neural Attention based Recommender System

  • Convolutional

    Recommender system on

    review

    data to learn user & item

    embeddings.

  • Attention

    mechanism used to identify & weight vital users for a business and vice versa.
  • Achieved

    1.04 RMSE

    score on review rating prediction. Compared to baseline factorization.
    • Github

    Seq2Seq Abstractive Summarization

  • Generated abstractive text summaries for news articles in

    Daily Mail dataset

    , and achieved

    ROUGE-1 score of 24.1

    .
  • Constructed Seq2seq model with Bi-LSTM layer encoder and LSTM layer decoder with

    attention mechanism

    .
    • Github

    Voice Style Transfer

  • Understood the limitations of past generative approaches, like Variational Auto Encoder(VAE) and CycleGAN.
  • Our modified

    StarGAN

    works on unpaired audio data and is capable of

    multi-speaker Voice Conversion

    .
    • Github

    NBA Outcome Predictor

  • Scraped data from websites and visualized it to make features to predict match winner based on past data
  • Evaluated models like Random Forest, XGBoost.

    Support Vector Machines

    gave

    F1 score 0.72

    on '17 season
    • Github

    Contact

    Phone Number:

    +1 812 606 9613

    Email:
  • dhruuv111@gmail.com
  • dagarwa@iu.edu