Download PDF

Technical Portfolio

Academic Background

Aug 2013Apr 2017

B.E Computer Science and Engineering

S.S.N College Of Engineering (Affiliated to Anna University)

CGPA - 8.76 (At the end of 4th Semester)

[Currently in 6th semester]

Related Coursework:

  • Programming And Data Structures - I,II
  • Design and Analysis Of Algorithms
  • Probability and Queueing Theory, Discrete Mathematics
  • Database Management Systems
  • CS224d: Deep Learning For Natural Language Processing (External Coursework, Citing Externally from Stanford University)
  • CS231n: Convolutional Neural Networks For Visual Recognition (External Coursework, Citing Externally from Stanford University)


Jun 1999Jun 2013

High School (CBSE - Central Board Of Secondary Education)

Bala Vidya Mandir

CGPA - 9.8 (10th Grade)

Percentage - 96.6% (12th Grade)

Certifications

Academic, Research and Industrial Experience

Nov 2015Feb 2016

Research Engineer

Snapshopr

Projects and Experiments:

  • Text + Visual Sentiment Analysis
    • Identifying sentiment of data which involves both image and text such as Twitter, Snapchat. 
    • Visual sentiment is calculated by identifying ADP (Adjective Noun Pairs) associated with each image using Convolutional Neural Networks.
    • Text sentiment is calculated using a simple Bag-Of-Words model.
    • Dataset : Visual Sentiment Ontology Dataset, Twitter 
  • Image Captioning
    • Automatically describing content of images using Multimodal Recurrent Neural Networks (m-RNN) and Convolutional Neural Network features.
    • Currently ranked 14 (BLEU-1 C-5 metric) in CodaLab leaderboard. 
    • Dataset : MSCOCO - Microsoft Common Image In Context.
  • Character Level Language Modeling
    • Trying to generate new text, character by character using character level language models.
    • Language Models used : Recurrent Neural Networks, Long Short-Term Memory Networks (LSTMs).
    • Dataset : Rap lyrics scraped from public data
  • Visual Question Answering
    • Creating a model that answers questions regarding Object, Number, Color, Location given an input image.
    • Model was created using Convolutional Neural Network features and BOW features for the question, MLP to perform question answering.
    • Dataset : Toronto COCO-QA Dataset
  • Question Answering Of Simple bAbI Tasks
    • Performing question answering of simple tasks such as factoid QA with one/two/three supporting facts, argument questions, etc.
    • Method Used : Memory Networks , End-to-End Memory Networks
  • Painting Style Classification (Snapshopr Hackathon)
    • Classifying paintings by style using Convolutional Neural Network features and gradient boosted trees classifier
    • Dataset : Scraped from WikiArt
  • Yelp Restaurant Photo Classification (Kaggle Problem)
    • Tagging yelp restaurant photos with 8 independent attributes using convolutional neural network features and XGBoost (EXtreme Gradient Boosted Trees) classifier.
    • Current ranked 8 on the leaderboard.
  • Object Recognition and Detection for Fashion Domain and Visual Product Recommendation
    • Recognizing and localizing clothing items such as dress, tops, skirts using R-CNN (Regional Convolutional Neural Networks) 
    • Visual product recommendation is performed using distance metric learning techniques such as Siamese Networks (LeCun et al.), Deep Ranking.
    • Saliency Detection techniques such as DeepDetect (CVPR 2015) are used for image preprocessing.
  • Programming an AI to learn to play Ping Pong using Deep Reinforcement Learning
    • Teaching an AI agent to play Ping Pong using Deep Q-Learning method.
    • The algorithm receives only image of game state and reward for each action as input.
    • Ping Pong game - developed using PyGame
  • Tools learnt/used : Theano, Keras, Caffe, GraphLab
Jun 2015Jul 2015

Student Intern

Pith Inc.
  • ​ Visualizing data from raw Bluetooth Low Energy sensor signals
    • Identifying the breathing signals
    • Analyzing the effect of interpolation of breathing signals from multiple  sensors
    • Analysing breathing patterns and designing a peak detection algorithm
    • Tools used:
      • Python, Matplotlib, R
  • Designing a fast,efficient  script runnable on a server to scrape public medical data
    • Parsing/Processing the scraped data and extracting useful information
    • Designing a DB and pushing the data into the DB using python-sql connect.
    • Tools Used:
      • Python, BeautifulSoup, Mechanize, Sqlite, Linux Shell Scripting
Mar 2015May 2015

Data Science Capstone

Johns Hopkins University - Partnered With Swiftkey Corporation at Coursera
  • Designing a Next-Word Prediction App that is fast, accurate, responsive and deployable online at shinyapps.io
    • Skills Learnt:
      • Text Preprocessing (Data Cleaning, Tokenization, Sentence Segmentation)
      • Text Mining, Exploratory Analysis and Data Visualization
      • Language Modelling
    • Tools Used:
      • Shell Scripts, R and RStudio, Shiny (Creating web-apps)
  • Project Presentation : Data Science Capstone Project
  • Project Report : Data Science Capstone Milestone Report
  • Project Display : Next-Word Prediction App

Skills

Python

Scripting wide variety of tasks such as web scraping, machine learning, algorithms programming, etc. using Python.

Algorithms

Performing algorithms programming and efficient data structures programming using C, C++ and Python

Machine Learning & Data Science

Programming machine learning and Data Science using Python and Statistical Programming using R.

Shell Scripting

Using the full power of the Linux Shell to perform tasks at very high speed using shell commands and shell scripting.

Deep Learning

Performing Natural Language Processing, Computer Vision using Deep Learning techniques and frameworks such as Theano, Keras

Test Scores

Graduate Record Examination:

  • Overall Score: 329/340
  • Quantitative Reasoning Score : 170/170 (98 Percentile)
  • Verbal Reasoning Score : 159/170 (81 Percentile)
  • AWA : 5.0/6.0 (93 Percentile)