About

I am a Software Developer at VMware and have experience developing microservices, databases, DevOps, automation, and analytics.

I completed my Bachelor's (B.E.) in Computer Engineering at Sardar Patel Institute of Technology, Mumbai, India, in May 2020.

I am immensely interested in AI Research and have authored a couple of papers. Moreover, I have won a couple of hackathons themed more or less on AI.

I blog about Machine Learning, Deep Learning, Reinforcement Learning, Natural Language Processing at Medium.

Please have a look at my resume for more information.

Publications

My Google Scholar Profile

2020

Jagtap. et. al. (2020). Healthcare Conversational Chatbot for Medical Diagnosis. In Patil, B., & Vohra, M. Handbook of Research on Engineering, Business, and Healthcare Applications of Data Science and Analytics. IGI Global. http://doi:10.4018/978-1-7998-3053-5. [Webpage]

Jagtap, Rohan, and Sudhir N. Dhage. "An In-depth Walkthrough on Evolution of Neural Machine Translation." arXiv e-prints (2020): arXiv-2004.[Paper]

Hackathons

Winner of Smart India Hackathon Software Edition (March 2019)

Insurance companies often have communication issues with the customers as they seldom bother to read large policy documents, hence resulting in conflicts. The task was to provide a technological solution to this communication gap.

The proposed solution is to develop an Extractive Text Summarizer with an additional translation and text-to-speech support for the policy documents. I contributed to the core summarizer logic by implementing TextRank and Term Frequency and integrated it with the rest of the components using the Django MVC framework.

Check out the code here.

Runner Up at Sangam - ML Hackathon by IIT Madras (Aug 2019)

Competed with 100+ teams on the online event and got selected among the top 25 who were invited at the onsite event, which was held at IIT Madras.

The task was to derive Air Quality Index (AQI) and valuable insights from the given data and to predict AQI in Temporal as well as Spatial dimensions with supporting high-quality visualizations.

We proposed training two separate autoregressive models (ConvLSTM) for temporal and spatial predictions, respectively. I took care of pre-processing and training the models, whereas my partner, Siddhesh Jadhav, worked on the tableau software for the visualizations.

Check out the code and visualizations here.

Top Articles

T5: Text-To-Text Transfer Transformer

Understanding Transformer-Based Self-Supervised Architectures

In this article, we’ll discuss Google’s state of the art, T5 — Text-to-Text Transfer Transformer Model which was proposed earlier this year in the paper, “Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer”.

Reformer: The Efficient Transformer

Understanding Transformer-Based Self-Supervised Architectures

Abstractive Text Summarization Using Transformers

An exhaustive explanation of Google’s Transformer model; from theory to implementation

This article is a step-by-step guide for building an Abstractive Text Summarizer for generating news article headlines using the Transformer model with TensorFlow.

Transformers Explained

An exhaustive explanation of Google’s Transformer model; from theory to implementation

Understanding Markov Decision Process (MDP)

Towards Training Better Reinforcement Learning Agents

In this article, we’ll be discussing the objective using which most of the Reinforcement Learning (RL) problems can be addressed— a Markov Decision Process (MDP) is a mathematical framework used for modeling decision-making problems where the outcomes are partly random and partly controllable.

Projects

Abstractive Text Summarization using TensorFlow

Implementation of Transformer (Vaswani et. al.)

Autoregressive Language Model-I: Book Text Generation

Implementing a RNN-based Language Model from Scratch using Numpy

Autoregressive Language Model-II: Eminem Lyrics Generation

Implementating a LSTM-based Language Model using Keras

MNIST Digit Generation with GANs

Implementation of DCGAN: Deep Convolutional Generative Adversarial Networks (Radford et. al.)

Constructing Images from Wireframes with pix2pix

Implementation of pix2pix (Isola et. al.)

The pix2pix generator is based on the U-Net architecture. To prevent loss of information from the generator bottleneck, U-Net proposes corresponding skip connections from encoder to decoder.

The authors have proposed an enhanced classifier for the discriminator. It essentially classifies patches of the original image as real or fake to capture high-frequency structures. It is called Markovian Discriminator or PatchGAN.

Rohan Jagtap