Animesh

Animesh

Machine Learning Engineer

Professional Experience


Wynk Limited

Machine Learning Engineer
08/2020 - present | Gurgaon, India

RNN-based Recommendation System

  • Developed an RNN-based recommendation system, that takes in a sequence of songs as its input, and predicts the next most suitable song that is most likely to be consumed by a user, in continuation with the previously played songs.
  • Resolved recommender system’s common problem known as the Popularity Bias.
  • Explored ways to train this neural network-based model, on multiple GPUs (distributed training). This allowed us to train our model on bigger data, in less time, making it possible to perform more experiments in the same amount of time.

Personalized Playlist Recommendation

  • Built a PySpark-based pipeline for preprocessing the data required for generating personalized playlists recommendations.
  • Developed a methodology that takes in the precomputed embeddings of users and songs as inputs and generates 20 personalized playlists recommendations for every user such that all the recommended playlists are aligned with the user’s taste, and there exists diversity in playlists languages.

UnitedHealth Group

Data Scientist
04/2019 – present | Gurgaon, India

Compassionate Letter Generation

  • The objective of this project was to find sentiments in Insurance Claim Denial Letters and transforming the letters with low compassion scores into letters with improved compassion scores.
  • Extracted the sentiments from the Insurance Claim Denial Letters by using DeepMoji, an attention-based residual text classifier that is capable of generating a rich representation of a given input text and also produces probability scores for 64 emojis.
  • Transforming a text with a low compassion score into a new text with a higher compassion score by applying text style transfer, but at the same time keeping the original content of the text intact. (This step was still in progress when I left the organization)

Document Form Digitization

  • The objective of this project was to digitize the documents in the healthcare domain, segment all the fields present in the document form, and then extract the text as a key-value pair.
  • Extracted text using Multi-Scale Template Match Algorithm
  • Noise Removal using UNet-based Denoising AutoEncoder Printed Text Recognition

Tata Consultancy Services

Developer
11/2017 – 03/2019 | Gurgaon, India

Worked on an Industrial Document Analysis project in Innovation Labs, TCS where the objective was to localize and perform recognition of hand-written text on document images.
Key Contributions-

  • Text extraction using CTPN (Connectionist Text Proposal Network)
  • Noise removal using Fully Connected Network
  • Segmentation of the individual characters from the extracted text patches using Connected Components Algorithm
  • Text Recognition using CNN and Capsule Networks

Tata Consultancy Services

Assistant Systems Engineer
03/2017 – 10/2017 | Gurgaon, India

Built pipeline for a Computer Vision project in Python which included writing preprocessing and postprocessing functions using OpenCV and made a basic GUI in Tkinter to demonstrate the automation capability.