Abhijit Kumar

About Abhijit Kumar

My core competencies include tweaking and implementing various supervised machine learning algorithms to drive data driven decision making. Since past few years, my focus area has been natural language understanding specifically devising attention or recurrence mechanisms for multimodal data in a multi-task environment.

Experience

1. Applied Scientist Ii

   Amazon

   Jan 2019 — Present

   Working in procurement space of Amazon. The work primarily revolves tweaking attention mechanism in transformer network to propel conjugated learning approaches for multi-modal data in a multi-task setting. The downstream task ranges from product classification to product matching and recommendation. •Worked on optimizing search complexity and finally guiding efficient product recommendations for amazon sourcing website. With an estimated saving of $10M that would have been lost due to leakages. •Worked on causal machine learning problems to attribute spending for billions of primecustomers worldwide. In summary, worked to measure renewal lifts caused by variousprime benefits. Thereby attributing causal lifts in renewal of prime members to multiplebenefit usages. •The prototype using debiased machine learning, is based on the Rubin’s potentialframework, where the incremental effect of a treatment class is studied by comparing itagainst the behavior of a reference population.

Education

• Purdue University

  Master's degree, Quantitative Methods

• Manipal Institute of Technology

  Bachelor’s Degree, Electrical and Electronics Engineering

• Purdue University

  Master's degree, Quant