Experience

• 

  Explara

  Dec 2014 - Jan 2015

  Programmer

  • Worked on automated text classification at Explara.com.• Implemented a Python application to read event descriptions at classify them into sets of categories.• The program runs as a cron job that reads bulk event categories written in natural language and tags them with appropriate categories.

• 

  Uc santa cruz + stanford aspiring researchers' challenge

  Jan 2015 - Jun 2015

  Research associate

  I participated in a mass research program initiated by UC Santa Cruz and Stanford University, to study the effectiveness of crowdsourcing. I investigated the variation of cost with accuracy of human workers when multiple crowdsourcing tasks are displayed as one task to the worker. The studies carried out in this project were based on human workers aiding computer vision algorithms.

• 

  Stride.ai inc

  Jun 2015 - Jul 2015

  Programmer

  • Worked on text processing with Python at Tationem.com (now Stride.ai)• Developed a program for anaphora resolution and subjective descriptor mapping. The program uses SpaCy and WordNet to infer count and gender features of nouns and maps pronouns to entities in the text. It also infers features of entities based on the matched pronouns and their usage in the text.• The algorithm I implemented is being used to develop a subjective sentiment analysis API.

• 

  Intel corporation

  Jun 2016 - Jul 2016

  Intern

  • Researched techniques to automatically predict the passenger’s perception of the drive quality in different types of vehicles.• Designed experiments and developed apps for data collection.• Developed novel approaches for preprocessing accelerometer data to train convolutional neural network (CNN) models. Dataset augmentation and preprocessing steps ensure that the model learned by the CNN is invariant to the angle and in-vehicle position of the accelerometer.• Trained TensorFlow CNN models to identify bad driving events and anomalous road surfaces. Show less

• 

  Intel corporation

  Dec 2016 - Jun 2017

  Intern

  Hyperspectral Imaging:• Researched different convolutional neural network architectures (TensorFlow) for classification of hyperspectral images. Used to neural networks for semantic segmentation of these images of different food and substances with similar visual features.• Developed a toolkit to use for real-time, non-destructive identification of food grain quality from hyperspectral images of grain samples.Voice analysis• Researched various features currently used for sound and voice analysis with machine learning.• Surveyed historical and state-of-the-art methods (hand-crafted features and end-to-end models) used for emotion recognition from voice. Show less

• 

  Parc, a xerox company

  May 2018 - Aug 2018

  Intern

  • Techniques for few shot learning.• Sample efficient object detector training.

• 

  Amazon

  Jul 2019 - now

  • Amazon Rufus (https://www.amazon.com/Rufus/). • Amazon Rufus.• Large scale distributed training. Training LLMs from scratch.• Inference with custom hardware. Optimizations for low-latency, high-throughput inference.• Controllable text generation.• Release announcement -- https://www.aboutamazon.com/news/retail/amazon-rufus• Release video -- https://www.youtube.com/watch?v=uXkDMBNCoFo• See also -- https://spectrum.ieee.org/amazon-rufus • Large scale multilingual, multimodal model training. • Large scale multilingual, multimodal model training. • DashCart Project: https://www.amazon.com/b?node=21289116011• On-device machine learning: sensor processing, navigation, computer vision.

  • Senior Applied Scientist

    Sept 2024 - now

  • Applied Scientist II

    Jan 2023 - Sept 2024

  • Applied Scientist II

    Apr 2022 - Dec 2022

  • Applied Scientist

    Jun 2021 - Apr 2022

  • Applied Scientist

    Jul 2019 - Jun 2021