Alexa International

Worked on instrumentation of cloud infrastructure for Samsung Bixby Smart Assistant by integrating it with Dynatrace, performing function call chain analysis, with CPU, memory and network usage division among different components such as speech recognition, NLP, text-to-speech and natural language generation.

• Developed Online machine learning topologies for regression and classification using Node.js for IOT devices.• Designed and implemented an O(logn) algorithm for delivering software updates to 6000 machines by employing a binomially growing P2P file distribution strategy and using Kafka as a distributed communication system.

• Contributed new features and bugfixes to the web dashboard, user interaction and cluster graphs of Apache Flink.• Developed a machine learning library for Apache Flink with distributed algorithms for linear regression, decision tree, k-means++ and k-means++ parallel clustering, online histograms and data statistics.

• Implemented custom distributed k-means clustering for Apache Mahout, with serialization using LZO compression and map-reduce-combine, attaining a speedup of 6x in preprocessing and 2x in per-iteration time.• Created a distributed algorithm for training Random Forests on Hadoop using map-reduce-combine operations to evaluate splits across all tree nodes, which scaled upto 20GB data on an 8 machine EC-2 cluster.