Abhijeet Chowdhury

Experience

• 

  Indian Oil Corporation Limited

  Jun 2016 - Jul 2016

  Summer Intern
• 

  University of Regensburg

  Jun 2017 - Dec 2017

  Intern
• 

  ZS

  Jul 2018 - Aug 2019

  Associate

  Significantly optimized a client's Risk Data Library by implementing various enhancements, leading to a 7% reduction in man-hours within the fraud claims department. Utilized C++, CMake, and Git for development and version controlIntegrated external data sources using gRPC to support real-time fraud analysis pipelines

• 

  American Express

  Sept 2019 - Dec 2022

  Utilized deep learning, supervised learning and NLP to optimize the new acquisition journey of digital users. Some projects:Leveraged GRU-based (LSTM like) RNNs to model user interactions and conversions, then used SHAP to understand feature importance, guiding our A/B testing backlog. Approximately 50% of the hypotheses generated were successfully implemented after A/B testing.Optimized ML model training for a CMS-based card recommendation system, resulting in a 7% increase in CTR post-deployment by aligning recommendations with user spend profilesExplored LSTM and GRU architectures with embedding layers on Amex chat data to identify user pain points. Ultimately, TF-IDF with Logistic Regression proved more efficient, and implementing it revealed pain points impacting 10% of annual shop revenueCreated Big Data pipelines to process user digital fingerprint into data layers fit to be consumed by ML pipelinesConducted multiple A/B tests on Amex’s commercial assets. Tests ranged from hypothesis testing of a new feature to a new ML model deployment for better interaction or conversion. Show less Worked mostly on hypothesis generation using machine learning and testing them out using A/B tests. Also got experience on a real time inference system from development to deployment.Conducted several A/B tests to validate running hypothesis and optimize user digital experience.Implemented an operational data layer which summarized user digital fingerprints using Hive SQL which reduced average A/B test read time from 4 days to a few hours.Prototyped a hybrid bot detection model using GRU networks (LSTM like) and Autoencoders to analyze user behavior. Preprocessed terabytes of data with Dask to extract behavioral patterns. Optimized the model to capture both temporal dependencies and anomalous behavior, achieving a F1 score of 89% on test dataset. Worked with Cloud Team to deploy above bot detection model in real-time. Optimized further for AWS deployment using EMR and SageMaker Show less

  • Data Scientist (Big Data)

    Mar 2022 - Dec 2022

  • Senior Associate

    May 2021 - Mar 2022

  • Associate

    Sept 2019 - Apr 2021
• 

  Khoury College of Computer Sciences

  May 2023 - now

  - Working under the guidance of Dr. Mike Shah (Yale Graphics) and Dr. Crane Chen (Disney)- Researching ways to improve isosurface rendering for AR/VR usecase- Additionally, worked on engineering of an image manipulator based on tiling theory - sponsored by Adobe Research. The paper is submitted for SIGGRAPH 2025- Currently being sponsored by Khoury Apprenticeship Spring 2025 and Adobe Research Sponsored Project Fund Teaching Assistant for Computer Graphics (Boston - Dr. Mike Shah and Silicon Valley Campus - Dr Crane He Chen) Intensive Foundations of Computer Science Data Structure and Algorithms

  • Research Assistant

    Jan 2024 - now

  • Graduate Teaching Assistant

    Sept 2024 - now

  • Graduate Teaching Assistant

    Sept 2023 - Dec 2023

  • Graduate Teaching Assistant

    May 2023 - Aug 2023
• 

  Storm Flag Games

  Jan 2024 - May 2024

  Associate C++ Engineer (Graphics Engineer)

  Implemented Environment Effects like Fog, Bloom, Skybox, and Volumetric EffectsgRPC, Cache Optimization and Load BalancingRendering pipeline debugging and optimizationGameplay Development

• 

  Intuitive

  May 2024 - Aug 2024

  Software Navigation Intern (Graphics Engineer)

  Optimized lung scan transparency rendering for Ion's GUI during biopsy, implemented Depth Peeling with layer offsettingImplemented a GPU based collision and subdivision scheme for non-depth tested workflow to reduce transparency artifacts