Arun Rajendran

rajendranarun95@gmail.com

I am a Machine Learning Engineer with 3.5 years of Machine learning experience specialized in NLP. I have a Applied Mathematics Masters degree from University of British Columbia. I am really passionate about data science and machine learning. I have experience in NLP, Computer Vision, Deep learning, High-performance computing, machine learning and MLOps.

Recent News

  • Jan 20, 2023: Lost job as part of company wide layoff at SoundHound
  • Apr 15, 2022: Joined SoundHound Technologies Inc. as ML Engineer II
  • Mar 31, 2022: Quit PatriotOne Technologies Inc.
  • Oct 10, 2021: Promoted as Lead Machine Learning Engineer at PatriotOne Technologies Inc.
  • Sep 15, 2021: Joined as Machine learning Engineer at Xtract AI, A PatriotOne Technologies Company

Experience

Machine Learning Engineer II, SoundHound Technologies Inc.

  • Created a paraphrasing application with a customized frontend and Kubernetes-based backend for synthetic data generation
  • Trained DistilBert for identifying incorrect input queries & reduced 10% of processing time in production
  • Developed & Deployed entity tagger application with multiple methods & Implemented unit testing
  • Setup monitoring and alert system for all the deployed applications using Jenkins to reduce downtime
  • Analyzed production data to identify issues & developed solution modules within the post-processing pipeline
  • Trained Pegasus model for transforming badly-formed conversational user queries & voice commands

April 2022 - Jan 2023

Lead Machine Learning Engineer, PatriotOne Technologies Inc.

  • Lead a 1M multi-person project & built an end-to-end smart information monitoring system with different forms of visualizations for news articles
  • Developed summarization models, entity linking knowledge graphs & implemented anomaly detection algorithms for extracting outlier articles
  • Understood the needs of stakeholders and iteratively designed wireframe templates for the product
  • Developed Image captioning models for alt-text generation, performed Image enhancement, OCR, and figures/tables were extracted from scanned documents
  • Implemented and maintained Jira Kanban boards to monitor and track project progress
  • Setup custom annotation server for specific entities extraction task with code reviews & testing for standardized code & documentation.
  • Collaborated on open-source python package development for training & incorporated ML life-cycle management using MLflow

Oct 2021 - Mar 2022

Machine Learning Engineer, Xtract Technologies Inc.

  • Fine-tuned BERT model for gender prediction from text and username using for Reddit and Twitter
  • Trained BERT model to identify potential threats using GDELT 2.0 and linked it to social media
  • Models for figure/table & sequence extraction from healthcare papers were deployed in AWS services
  • Established scalable ML detection pipelines including EDA, data imputation modules & feature engineering algorithms for identifying donor-recipient matches
  • Lead the project from inception to completion by designing product workflow, training models & creating microservices (Docker/FastAPI)
  • Beautiful-soup based web crawler was created to scrape multiple web pages and standardize them
  • Trained GPT-3 & Bert models for NER, document classification, and comprehension based question answering tasks
  • Custom models were devised for semantic segmentation, fight classification & crowd anomaly detection using Transfer learning
  • Produced Baseline models for vehicle make-model identification & vehicle re-identification
  • Iteratively built an interactive dashboard using Plotly dash for YOLO object detection on the video feed

Aug 2019 - Sep 2021

Data Analytics Intern, Spini

  • Analyzed and visualized data using Tableau. Created tags using google tag manager to track user interactions
  • Developed a wide and deep model in Tensorflow with 99.4% accuracy that improved the lead conversion by 20\%

May 2017 - Jun 2017

Data Analytics Intern, FastNext

  • Developed a set matching algorithm with elasticsearch and kibana to obtain patterns in data
  • Performed data cleaning with R and compared different machine learning models using python's scikit library

July 2017 - Aug 2017

Super Coordinator, Institute Website Operations, IIT Madras

  • Independently trained and led a team of 5 to develop the Information portal using AngularJS, SQL & PHP
  • Revamped the Landing page of Institute website to make it more interactive and user-friendly
  • Integrated Vanilla Forums into MiTr Website using Single Sign On that significantly improved discussion among students

May 2015 - Apr 2016

Head, Website Operations, Entrepreneurship Cell

  • Created the website of E-Cell 2016 and handled E-Summit registrations using SQL & PHP
  • Incorporated google analytics, used the data generated and improved the website increasing the traffic by 80\%

Aug 2015 - Apr 2016

Technical Projects

3D Parallel Navier Stokes Solver

Master Thesis, UBC

  • Developed efficient 3D,multi component Navier stokes solver using Direction Splitting
  • Implemented parallelization using domain decomposition through optimized MPI communications
  • Studied schlur complement technique,uzawa algorithm & lagrange multiplier for domain decomposition

Sentence-Level BERT & Multi-Task Learning of Age & Gender in Social Media

Submitted to CONLL 2019

  • Tested variations of Multitask BiGRU (task specific & shared attention) for predicting age-gender simultaneously
  • Compared performance with single task models & small BiGRU models & analyzed the words attended by the model
  • Evaluated bert model performance on individual tasks & analyzed sequence length model performance

BERT and Hierarchical Attention Multi-Task Learning of Fine-Grained Dialect

Submitted to EMNLP 2019

  • Pretrained BERT model from scratch on arabic data & fine tuned on development & test set to report performance
  • Exploited variations of multitask learning(hierarchical & shared model) for city,state & country level dialect prediction
  • Tested the performance of self attention with supervised attention(using mutual information gain)

Ensemble Learning of Offensive Content With Enhanced Training Data

Github Paper
  • Examined several traditional machine learning models and used majority class as baseline
  • Conducted hyperparameter optimization to identify best sampling technique and n-gram value
  • Performed synthetic data generation technique(word2vec aided paraphrasing) to obtain balanced dataset
  • Analyzed different ensembling methods to obtain 6th position (75 teams) & 9th position (65 teams) in subtasks B & C

Happy together: Learning and Understanding Appraisal From Natural Language

Github Paper
  • Carried out several traditional machine learning models and defined the best model as baseline
  • Explored variations of LSTM with different embeddings for sociality and agency classification on HappyDB
  • Visualized words attended by best model and ranked them to understand the model performance

Freelance Projects

Census Income Prediction

Github R Script

Exploited various sampling techniques and stacked multiple models in R

Bigmart Sales data

Github Jupyter Notebook

Identified feature relationships by visualising data using seaborn and created new features

Loan prediction

Github

Created model pipeline of cleaning,feature engineering,training and ensembling using makefile in R

Human Activity Recognition

Github Jupyter Notebook

Explored Hyperopt module for tuning model parameters to improve performance

Black Friday Problem

Github Jupyter Notebook

Utilized Bokeh library for visualizing data and performing feature selection/engineering

Bike demand Prediction

Github Jupyter Notebook

Performed data visualization using Plotly and Graphviz libraries

Wine quality dataset

Github HTML

Investigated different regression models in Matlab to predict wine quality

Student Evaluation dataset

Github HTML

Applied different classification models in Weka to predict number of retakes

Credit Card Fraud Detection

Github Jupyter Notebook

Applied different classification models in Weka to predict number of retakes

General Chatbot

Github Jupyter Notebook

Built a chatbot based on Cornell Movie Corpus using Attention mechanism and beam search decoding

Dog Breed Classifier

Github Jupyter Notebook

Trained CNN model with transfer learning(Inception V3) to classify dog breeds.

TV Script Generation

Github Jupyter Notebook

Created a word level LSTM language model in tensorflow on Simpsons dataset.

Neural Machine Translation

Github Jupyter Notebook

Tested a sequence to sequence model for language translation from english to french.

Face Generation

Github Jupyter Notebook

Generated faces (celebA dataset) and numbers (mnist dataset) using GAN's.

Facial Beauty Prediction

Github Jupyter Notebook

Implemented Resnet50 model(Transfer learning) in Keras to predict face rating.

Breast Cancer Classification

Github Script Paper

Identified the best classifier(Random forest) and used unsupervised dimension reduction techniques to visualize the dataset

Character Generation

Github Jupyter Notebook

Generate characters based on Pride and Prejudice book using a character level LSTM model

Face Recognition

Github Script

Haarcascades in OpenCV are utilized to detect eyes,mouth and smile. Tested the same in a live video stream

Object Detection

Github Script

We use SSD to perform object detection and test it in a video.

Image Generation

Github Script

We do Image generation using DCGAN and test it by training on CIFAR10 dataset

Education

University of British Columbia

Master of Science
Mathematics - Applied

Grade: 89%

August 2017 - August 2019

Indian Institute of Technology, Madras

Bachelor of Technology
Chemical Engineering

GPA: 8.69/10

August 2013 - May 2017

Skills

  • Languages
  • C++(4 years)
  • Python(4 years)
  • SQL(3 years)
  • R(2 years)
  • Julia(6 months)
  • Web Design
  • React
  • PHP
  • JavaScript
  • NodeJS
  • MySQL
  • Libraries
  • Pytorch
  • Tensorflow
  • Dask
  • NLTK
  • spaCy
  • Frameworks
  • MLflow
  • FastAPI
  • Docker
  • Flask
  • Plotly Dash
  • Simulation
  • Matlab
  • Mathematica
  • Octave
  • Aspenplus
  • Ansys

Publications

  • Rajendran,Zhang,Abdul-Mageed. Happy Together: Learning and Understanding Appraisal From Natural Language. Affective Content Analysis Workshop @ AAAI (AffCon2019), Honolulu, Hawaii (January 2019) Arxiv
  • Rajendran,Zhang,Abdul-Mageed. Ensemble Learning of Offensive Content With Enhanced Training Data. 13th International Workshop on Semantic Evaluation (SemEval), Minnesota (June 2019) Arxiv
  • Zhang,Rajendran,Abdul-Mageed. Hyperpartisan News Detection With Attention-Based Bi-LSTMs. 13th International Workshop on Semantic Evaluation (SemEval), Minnesota (June 2019) Aclweb

Presentations/Conferences

  • Presented 'XLNet: Generalized Autoregressive Pretraining for Language Understanding' paper in UBC
  • Attended AAAI 2019 conference and exhibited our best performing system on CL-Aff shared task
  • Presented 'Multi-Task Deep Neural Networks for Natural Language Understanding' paper in UBC
  • Gave a talk on 'Pre-training of Deep Bidirectional Transformers for Language Understanding' paper as part of Deep Learning Research course
  • Delivered presentation of 'Attention is all you need' paper in UBC NLP group