Responsible for the heaviest hit Tubi service "FeatureStore" serving as the backbone of the machine learning infrastructure. (Tech stack: Akka; Scala; Scylla)

Within a few months stabilized the service experiencing frequent service degradations and occasional meltdowns with latencies measured in seconds. Now it handles up to 300K requests/sec with ultra low latencies (<~15ms) under 2x load. These improvements came perfectly in time for the world cup Tubi proudly broadcasted without any technical hick-ups.

Hands-on, lead engineer in big data processing, performance engineering, analysis and machine learning. Processing measurements of 500 million unique viewers watching 150 billion streams per year with 1.5 trillion real-time transactions across more than 180 countries

Develop and enhance core AutoML engine where we automatically analyze customer data and extract thousands of features and feed to our model factory and then auto produce the best models leveraging the best features are presented to our customers - without writing any custom code..

• Develop operators for time oriented features, histogram, target encoding; distributed geolocation matching with dbscan etc.

• Improve the architecture for better performance and scalability, extensibility

Lead architect for Tibco's flagship Data Science platform, that is unique in its ability to deploy machine learning models across various execution environments such as relational systems, R, Spark, and Cloudera etc.

Define the architectural roadmap to upgrade it to work with Spark 2 with AKKA messaging as well as prepare it to run on the Cloud.

Change Healthcare processes ~80% of US healthcare data.

Reporting to the CTO and 2 SVPs, responsible for architecture of the data platform architecture addressing: acquisition, data processing, and publishing to consumers and analytics apps; porting the data platform to AWS.

Provide technical leadership to teams collaborating during AWS data lab summit.

Prototyped patient de-duplication POC using locality sensitive hashing algorithm from Spark ML which demonstrated 25x performance improvement over existing implementation at 1/10th hardware.

Developed a policy driven access management platform using micro-services architecture with AKKA to support 10B authorizations per day, leveraging AWS IAM.