Dwarkanath Prabhu

Work Experience

CMS Inc. is an event management company specializing in academic conferences

Sponsor Registration System

• Created customized sponsor registration systems in ASP with a SQL Server database for 3 conferences incorporating all client requirements
• Designed a visual booth selection interface using CSS Grid

Paper Management System

• Transitioned 15 year old code to assign reviewers to papers from PHP to Python
• Automated creation of post-conference proceedings as static HTML pages with full-text search across pdf files of presented papers using Python

Air Quality Monitoring (Data scraping)

• Scraped historical data of air quality from PurpleAir REST API for pollution monitors using Python (requests)
• Estimated risk to Asthma patients based on PM2.5 data and CDC reports

Skin Cancer Detection Using Deep Learning (Computer Vision)

• Analyzed a labeled dataset of 10,000+ skin images to classify into 7 types of diseases
• Achieved validation set multi-class precision of 87.5% - a 25% improvement over baseline by using:
  • ResNet50 pretrained on ImageNet in TensorFlow
  • Data Augmentation and Weighted Loss to fix class imbalance

Power Generation Capacity Planning – Conference Paper (Optimization)

• Modeled and solved a bi-objective optimization problem using linear programming to minimize water utilization and cost for power generation in Texas with CPLEX in C++ over the next 25 years
• Built an app using Shiny framework for R to display optimal solutions on a map
• Wrote a manual on using CPLEX Concert Technology for C++ that has been downloaded and used by over 1000 users

Aasaanjobs.com (now OLX People) is a job portal for entry-level jobs in India

• Performed attribution analysis using data from Google Analytics, Kochava and SQL
• Measured the effectiveness of marketing spending by running A/B tests on ads on Facebook and Google
• Drafted the annual marketing budget to allocate spending on different online platforms and personnel
• Achieved 20,000 app downloads in three months since launch at an average cost of $1 per install

Supply Chain Transformation for a leading footwear manufacturer (10-month project)

• Explored sales data to classify SKUs into 3 classes; Forecasted demand and formulated aggregate plans for each class
• Partnered with operations teams to identify opportunities and build solutions in the form of production schedule and procurement plan
• Improved order fulfillment from 65% to 90% and increased rated production capacity by 20%

Projects

Week In Memes is a blog that maintains a record of all Indian meme templates as they are generated and used

• Wrote 100+ articles in 1 year using a Python script that scrapes Twitter for memes on trending topics
• Designed a custom theme for Hugo (a Go based static site generator) to publish the blog
• Devised Python scripts for automation of social media posting to Twitter and Pinterest saving time by 80%
• Automated liking posts, commenting and following users on instagram; Increased followers by 300% in a week
• Grew to 10,000+ visitors/month using Search Engine Optimization, Email Marketing and Social Media at no cost

• Analyzed a data set of 76,000 drilling equipment failure events with 172 features
• Tested several models of SVM and Random Forest by tuning hyperparameters till the best validation set F1 score of 99% was achieved
• Deployed the best-performing model to a web app built in Flask that takes a csv file of features as input and predicts the probability of failure

• Collected data of almost 20,000 code files by scraping GitHub with its GraphQL API
• Built a data set of files labeled with the language each file is written in
• Trained LSTM model in Keras to predict whether file contains Java or not with 99.8% validation set accuracy

• Built a web app with Flask back-end, Bootstrap template and SQL database for user-generated posts like Twitter
• Implemented ElasticSearch to search for posts and users. Deployed to cloud via Heroku

• Conducted PCA on a manufacturing data set of 552 samples using Python to reduce dimensions from 209 to 4
• Applied quality control analysis with a threshold of 5% Type I error to remove out-of-control recursively data till only in-control data was left

Education