Apr 2023 – Mar 2024 · 11 mos · San Francisco, California, United States · Hybrid

May 2018 – Aug 2023 · 5 yrs 3 mos · On-site

• Initiated and co-founded Govoroon at Area120 (Google's internal startup incubator) which delivered Machine Translation solutions for customer support; authored and published a paper on neural machine translation; got 10 customer contracts signed within 8 months.
• Key contributor to AdLingo (Area 120) and Business Communications, focusing on APIs, measurement, reporting, and web applications. Built domain knowledge on the Ads ecosystem at Google as well as architecting systems for scale.
• Conducted ML training sessions for Googlers over 3 years, with 200+ Googlers attending in person. Also trained external community members from 15+ countries to become ML trainers within their own local communities.

Nov 2016 – May 2018 · 1 yr 6 mos · On-site

• Spam and abuse detection on Google Search

Jan 2016 – Jan 2016 · 0 mo

Jun 2014 – Jul 2016 · 2 yrs 1 mo · Bangalore

• Key projects:
• End-to-End Real Time Data Platform
• I helped design a Kappa architecture based model for the next generation Real Time Data Platform in [24]7. I also implemented a proof-of-concept of this - it uses a streaming engine to pre-compute session level variables and a graph database to store historical data, user profiles and link omni-channels.
• Technologies used: Kafka, Spark Streaming, Akka, Flink, Neo4j, OrientDB, Cassandra
• Prediction-as-a-Service (PraaS)
• This product automates the steps of the machine learning process – variable selection, binning, model training, simulation, monitoring and model deployment. I implemented the deployment module in which each step in the offline model building process has an equivalent representation (a javascript function) in real time. All of these functions then get stitched together to form a logical evaluation tree and this tree is published to a global repository from where the evaluator App can call it.
• Languages Used: Python, Javascript, PMML (Logistic Regression, Naïve Bayes, Decision Tree, Random Forest and SVM models are currently supported by PraaS)
• Deployment of Prediction Entities
• I created prediction entities (views and models) for five clients. These were deployed on [24]7’s Real Time Data Platform.
• Languages Used: Javascript, MVEL, Java
• Page Categorization Tool
• This is a data visualization tool that represents page categories as a sunburst. Users can interactively delete, combine and modify page categories on this, and the final result is stored as a javascript function that captures all the transformations the original URL goes through to form these customized page categories.
• Languages Used: Javascript (d3.js)
• Reporting Tool
• This is also a data visualization tool that I built using d3.js. This allows users to view the number of visitors, visits and chats on each page of the website by traversing through URLs represented as a tree.
• Languages Used: Python, javascript (d3.js)

Jul 2013 – Dec 2013 · 5 mos · Bangalore

• Key projects:
• Simulation and Modeling Ready Datasets
• I wrote a series of Python and HiveQL scripts that consume raw events from the Hadoop file system and session-ize, process and transform this data into variables that can be used directly by analysts to build models. The scripts ran as a daily job and the output tables were stored in a column-oriented SQL database.
• Technologies Used: Hadoop MapReduce, Hive
• Modeling Workbench
• This tool allows analysts to perform Exploratory Data Analyses on their datasets and view the results visually. I wrote some of the backend R codes for this.
• Language Used: R

May 2012 – Jul 2012 · 2 mos · Bangalore, India

• NGO meet and Exhibition
• I held an NGO meet to facilitate co-ordination among NGOs in Bangalore. This was attended by over 100 NGOs. I also helped organize an exhibition of traditional art work by students of the Foundation.
• Poster and Pamphlet Design
• I designed posters and pamphlets that were used to publicize the NGO meet and the exhibition, as well as brochures and flyers that were used to promote the Foundation, its vocational training center and the work that it does.