I'm was leading the development of Distributed Machine Learning offering called snapML for IBM Power Systems. This library provides high-speed distributed, GPU accelerated the implementation of popular machine learning models on modern CPU/GPU computing systems and can be used to train models at wire-speed (as fast as the network can support) as new data becomes available. It improves training time up to 46x on the previously published benchmark. The details are available in the below link.https://www.ibm.com/blogs/research/2018/03/machine-learning-benchmark/Prior to this, I've enabled and optimized Apache Spark/Hadoop Ecosystem for Power Server. Introduced an open source add-on library called GPUEnabler written in Scala to ease the exploitation of GPU for Spark Machine Learning workloads. This package extends Spark’s whole- stage-codegen framework to automatically generate Java code to call user-written CUDA kernels and provide awareness of GPU Memory to Spark.I've also worked with NVIDIA development team to enable NVIDIA GPU and CUDA software stack on Power. As a part of this work I've modified their GPU device driver, CUDA Software Stack, CUDA User Mode driver to use power specific APIs and Linux VMM to make CUDA 5.5 generally available for IBM Power8 platform. This work provided the required foundation for IBM to build the fastest super computer named summit https://en.wikipedia.org/wiki/Summit_(supercomputer)

I was a Technical Lead for Cluster Aware AIX (CAA) project during this period to make the AIX kernel as cluster aware kernel to provide storage/network device monitoring, fault tolerant communication and cluster-wide event notification using the Autonomic Health Advisor File System (AHAFS). IBM High Availability Multiprocessing and Virtual I/O Shared Storage Cluster consume these services. This whole cluster management(CAA) would run at interrupt level to instantly detect the failure of a node or any of its services like network or storage in millisecond latency.

During this period, I worked as a Technical Chief Engineering Manager (TCEM/Architect) for AIX Base OS Development and Performance in both IBM US and IBM India. Some of the key deliverables handled are given below,• SME for malloc subsystem, introduced new allocators, lock free thread level buffer management, various performance improvement for malloc subsystem to sustain #1 spark for SPECint benchmarks for AIX and provide out of box performance for memory intensive workloads.• Optimized interrupt handler of user to kernel mode switch to speed system call performance.• Introduced Decimal Floating Point, Olson Time zone, provided consultation to many customers on debugging memory leak, performance optimization for AIX• Lead/Part of the team implementing key features like Active Memory Expansion, Dynamic System Optimizer, Virtualization Support, Security - RBAC etc..

As a Developer at oracle E-Business Suite development team, I’ve developed the front-end pages and business Objects for the pages using Oracle’s OA Framework which is based on J2EE

As a developer in this project I’ve developed numerous transaction modules using java-servlets, JDBC, HTML and JavaScript. I’ve also design and developed a complex printing engine for the batch account statement module using Pro * C.