Mar 2016 – Feb 2019 · 2 yrs 11 mos

• Working as the Team Lead for ML Team - handling machine learning in reviews, ratings and other products.
• Developed a recommendation system using factorization machine. Dockerized the model and provided gRPC apis to handle production load. Deployed the models using Elastic Beanstalk.
• Developed an image aesthetic quality scoring model by creating ensemble of linear and deep learning models. Wrote fisher vector implementation from scratch for linear model and created a spatial pyramid pooling layer for deep learning model. Fully dockerized this model for production deployment [2 months]
• Built a deep learning model based on alexnet trained on in-house dataset to classify images uploaded on Zomato into food, ambiance, menu & human category. Built apis to allow multiple products to utilize the model. Built robust data pipeline to classify images from multiple sources. [3 months]
• Generated review highlights using Natural Language Processing, with a test coverage of 75%+. Deployed web apps on a new subdomain behind an ELB. Automated the entire production process using Ansible. Published a research paper in ACM on a hybrid model for rule selection in aspect extraction. [6 months]
• Developed new scalable user credibility and restaurant rating algorithm [1 month]
• Developed a scalable review spam detection algorithm using tfidf, cosine distance, mini-batch kmeans clustering,
• gensim etc. [2 months]
• Developed the common base module to support other ML projects by wrapping services like redis, mysql, etc.

Jul 2015 – Mar 2016 · 8 mos

• Handled entire Zomato For Business Android (working end to end - api to apps)
• Published 20+ releases of Business app (including complete makeover in terms of design)
• Spearheaded optimisations for the core modules used in all Android apps (HTTP, serialisation, caching techniques etc)
• Optimized Network layer resulting in 30% overall speed improvement
• Reduced cold start time in Android apps by 300%
• Reduced size of Business app by ~40%
• Increased product robustness by putting additional focus on unit tests (junit/robolectric)
• Introduced automated functional tests for faster releases (espresso)
• Introduced and setup continuous integration for multiple Android teams (daily builds with jenkins)
• Reduced 70+ man hours from every release with tests and automation
• Handled backend apis for Business apps

Jan 2015 – May 2015 · 4 mos · Gurgaon, India

• Project Goals:
• 1. Build an android application which controls the functioning of Germ-Zapping Robots.
• 2. Build a web based system which takes care of all the hospital management functionalities, client management functionalities, device management functionalities, etc.
• 3. Create REST API to interact with other third party systems.
• See project description for more details.

Jul 2014 – Jan 2015 · 6 mos · http://peplet.com

• Peplet Inc. products increase productivity by automating monotonous chores. We built several apps including a fairly successful music controller.

Jun 2014 – Jan 2015 · 7 mos · Mumbai Area, India

• Developed a machine learning algorithm to automate the detection of a public transport boarding event for communication assisted road transport system.

Jan 2014 – Apr 2015 · 1 yr 3 mos · http://github.com/DevelopFreedom

• DevelopFreedom is an open source initiative to help budding developers learn and create useful tools along the way

Nov 2013 – Jan 2014 · 2 mos

• Mentored 4 students in Google Code-In 2013 who made several contributions to KDE and helped create the new homepage for JuK music manager

Jan 2013 – Jun 2014 · 1 yr 5 mos

• KDE is an international team co-operating on development and distribution of Free, Open Source Software for desktop and portable computing. KDE community has developed a wide variety of applications for communication, work, education and entertainment.