Lead and manage the technology and product functionsAcquired by Delhivery.

Build, lead and manage the forecasting and optimization team at Adap.tv's India centerHands-on development of the forecasting & optimization modules of Adap.tv's platformDesign, review, improve and implement backend infrastructure and algorithms

Designed and developed software for forecasting traffic volumes for video ads. This was built using Adap.tv's RIBS2 framework in C.

Frintit was selected by Excelerate Labs, Chicago (ranked in the top accelerators by Forbes) from over 400 startups across the world. I was involved in pitching to potential clients for targeted advertising and marketing opportunities. My other responsibilities included the design and development of the server infrastructure, web API and web application, integrating with social networking platforms like Facebook and Instagram to customize user experience and improve targeting of advertising.

I worked at the Real Time Systems(Retis) Lab, Scuola Superiore Sant'Anna, Pisa on researching distributed systems and operating systems for many-core/multi-core architectures

I worked on device drivers, embedded software applications for optical telecom. equipment for SDH,SONET and OTN. Over my employment at Tejas, I developed frameworks for improving reliability and robustness of software during equipment upgrades and against failures. I also filed for two patents pertaining to connection configuration and protection schemes in OTN clouds.

I developed a method of using association-rule mining on records of past purchases of customers to determine preferences. Using this information, we generated new product and service bundles for individual customers given their preferences to help targeted marketing. The estimated conversion to a purchase in our simulation showed a 10% increase over current methods.

I worked on methods to forecast time series data to aid production capacity planning using Support Vector Machines and comparing these against classical approaches such as moving averages and weighted moving averages. We also designed a heuristic method to tune the Support Vector Machines in order to reduce the time taken to train and predict, and at the same time without significant loss in accuracy.