Jan 2019 – Jan 2023 · 4 yrs · Bengaluru, Karnataka, India

Jan 2015 – Jan 2018 · 3 yrs · Greater Bengaluru Area

• I worked as a Machine Learning Leader leading four teams on Machine Learning projects. I am actively involved in defining Machine Learning charter for the org and getting it executed.
• I am the first and the only Scientist Bar Raiser(BR) at Amazon in India and currently grooming scientists on raising the bar of Machine Learning Science and its applications.
• Leveraging Amazon NeuralMotor Deep Learning Framework, PySpark, Hadoop(MapReduce), Mahout, scikit, R, Matlab and Octave to build state-of-the-art Deep Learning and Machine Learning algorithms for Ecommerce
• Filed three patents one of which one has been classified as a trade secret and the other two are in the process of being filed with USPTO.
• Presented state of the art work on Machine Learning in Ecommerce at AMLC 2022
• Organized a Full-Day workshop on "Machine Learning and Reinforcement Learning" in Amazon Machine Learning Conference 2019
• Presented novel Deep Learning techniques at AMLC 2019
• Presented Deep Learning work through a talk at ARD Conference 2019
• Delivered a talk on "Machine Learning @ Amazon" in Grace Hoppers Conference attended by thousands of people from a variety of organizations
• Presented a paper in the field of NLP based on Deep Recurrent Networks in the Amazon Machine Learning Conference(Seattle) 2016
• Presented a paper on a novel Machine Learning Technique in Amazon Machine Learning Conference(Seattle) 2017
• Presented a novel Deep NLP technique at Amazon Machine Learning Conference(Seattle) 2018
• Conducted a 3-hour Tutorial on Machine learning in Advertising Machine Learning Workshop 2018
• I am a Day-1 Science Mentor

Jan 2012 – Jan 2013 · 1 yr · Bangalore

• Worked on Machine Learning techniques for Latency Improvement of the Detail Page.
• I developed a novel Machine Learning algorithm for Amazon.com Detail Page which significantly reduced the latency of the Detail page exceeding the team's goals for 3 years. This work gave us the ability to introduce a number of latency-intensive CX improvements which led to a huge positive impact on the Amazon GMS as measured by A/B Testing. This was possible without slowing down the Detail Page due to the reduction in latency by the ML model.
• Owned the Machine Learning aspects of this project and partnered with the engineering team in implementing the algorithm for real-time scoring.

Jan 2013 – Jan 2015 · 2 yrs

• Worked as a Senior Data Scientist in the Data Science team of Adobe Target, a suite of 4 solutions - Personalization, Recommendations, Search & Promote and Test & Target.
• Invented novel Personalization, Attribution and NLP algorithms using Spark, MLLib, MapReduce, StanfordNLP, Deep Learning, etc.
• Two patent applications filed with USPTO as a result of this work and business goals exceeded across Personalization and Recommendations of Adobe Target

Jan 2011 – Jan 2012 · 1 yr · New Delhi, India

• Worked on the projects - "WE DID IT"(Automatic Stutter Detection), VOCA(Voice Of Customer Analytics) and other speech/text related technologies including ATTILA - IBM Speech Recognition Tool under the prestigious "Blue Scholar Program".
• I received the prestigious "IBM PATENT AWARD" worth $1500 and got two patents successfully published by USPTO.
• Detailed Description :
• 1) "WE DID IT" - Wise ExtremeBlues' Disabilities - Intervention after Detection using Innovative Technologies : The project had two main goals - Stutter Disability Intervention and Deafness detection in infants. I owned the Stutter Disability Intervention and came up with a novel algorithm to automatically reconstruct Clean Speech Signal from the detected Stutter without losing its naturalness. This system was selected for demonstration in IEEE ASRU - 2011 held in Hawaii, USA.
• 2) VOCA - Voice Of Customer Analytics : I invented a novel algorithm to enhance the performance of IVOCA(IBM VOCA) and introducing a couple of critical features. I also came up with a novel Relational Clustering Algorithm which had the following improvements over existing methods:
• a) The new system was able to cluster more than one element (every element can have more than one dimension).
• b) Ability to use the relationship information among the entities involved.
• c) The performance was enhanced by a new Principal Component Analysis technique on the combined cluster-able data.
• d) Ability to extract the information about the relationship between different elements in the form of co-efficients for each pair.

Jan 2010 – Jan 2010 · 0 mo · New Delhi

• Worked on NLP, Machine Learning and Pattern Recognition.
• Worked on the Extreme Blue project "WE DID IT"
• on Machine Learning, Signal Processing and Pattern Recognition aspects. This project involved creating an automatic Stutter detection and intervention system.