Oct 2019 – Sep 2021 · 1 yr 11 mos · Cupertino, California

Oct 2014 – Oct 2016 · 2 yrs · San Francisco Bay Area

• Collaborating with data scientists and data engineers to develop deep learning ( text, image , video ) applications on Spark + GPU framework.
• Server performance platform engineer for applications on Spark and Hadoop frameworks
• Internal consultant for big data workloads on Hadoop and Spark framework
• Machine learning and other HPC workload optimization using Spark on CPU and CPU + GPU distributed platforms
• Work on marketing proof points for IBM big data solutions ( including Cloud ) and customer proof-of-concept
• Containerization for Cloud: Spark on Docker solution planning and server/cloud deployment design.

May 2013 – Sep 2014 · 1 yr 4 mos · Raleigh-Durham-Chapel Hill Area

• Engineer in Compilers team. Focused on different compilers analysis and comparisons, and take the effort towards LLVM improvements.

May 2012 – Aug 2012 · 3 mos · Raleigh-Durham-Chapel Hill Area

Apr 2012 – May 2013 · 1 yr 1 mo

• Motivation behind this work was to come up with an energy efficient and a reduced cost
• LED driver control system. This work encompassed digital control theory and embedded
• system concepts. LEDs are increasingly finding application in almost every sector, like
• consumer electronics (TV, phones), space lighting (cluster of LEDs) and indication, which
• require a durable, inexpensive and an efficient solution. The thesis build upon an existing
• closed loop control system developed utilizing the RL-78 microprocessor development board
• and a boost converter working in the continuous conduction mode (CCM). The primary focus
• was on cost reduction while maintaining/improving the efficiency, and thus, the discontinuous
• conduction mode (DCM) of operation is implemented and tested. DCM operation results in
• cost reduction of the system as only one MOSFET is required in the circuit. Dimming was
• controlled by exploiting the discontinuous current property of the inductor already present in
• the circuit, as opposed to a dedicated MOSFET required for dimming, through the PWM
• signal, in CCM. The efficiency of DCM and CCM are also discussed in this work.
• Comparison was also drawn in terms of the embedded software and the resources needed in
• both modes. It was proved that DCM is indeed more energy efficient than CCM.
• The Thesis attempted to quantify efficiency in terms of Lumens (luminous flux). A
• comparison is drawn with respect to lumens per input wattage used in CCM and DCM
• operations

Jan 2012 – Apr 2013 · 1 yr 3 mos

• Assisted in 3 courses related to Embedded development :
• Undergraduate level course ( ECE 306 - Basics in Embedded Development )
• Graduate level course ( ECE 561 - Embedded Systems )
• Advanced Graduate level Embedded course ( ECE 785 - Advanced Embedded Systems )

Aug 2011 – Jul 2013 · 1 yr 11 mos

• Worked on a multidisciplinary project that involves sensor integration to micro controllers and networking for transfer of data. Used PIC and Beagleboard (ARM)

Mar 2009 – Jun 2011 · 2 yrs 3 mos

Jul 2008 – Feb 2009 · 7 mos