Feb 2023 – Mar 2024 · 1 yr 1 mo · Bellevue, Washington, United States

• Commerce Systems
• Worked on the commerce platforms microservice team to help engineer a Charge Management System API in Java SpringBoot to enable the configuration of charges to apply automatically and accurately on quoted invoices to clients reducing human error and financial leakage of over 3 million dollars per year. Added functional and unit testing support for this API using JUnit.
• Implemented Flexport’s transit time calculation engine by building Ruby / Rails Active Record tables for procurement ingestion endpoints in GraphQL , and building on React.js front-end elements enhancing predicted transit time accuracy for over 1000
• monthly shipments.
• Gathered feedback from cross functional business stakeholders to design and implement an algorithm for deduplication of charges during the building of the Charge Management API response, improving user experience and reducing over-charging of customers by 15,000 dollars a month.
• Built internal testing tools and Datadog synthetic tests. Implemented caching for external dependency calls using the Caffeine library in Java, resulting in a 30% latency improvement.

May 2022 – Aug 2022 · 3 mos · Menlo Park, California, United States

• Optimized pipelines for running Open Pre-Trained Transformer models in Python/PyTorch on the Facebook AI Research (FAIR) Accel team.
• Improved evaluation performance by 20% by moving model loading code to GPU and creating virtual sub-shards of data, impacting the millions of people using Facebook's machine translation systems.
• Enabled Pytorch profiling to gain performance and memory insights while training, fine-tuning and evaluating large scale NLP models
• Contributed to the open-sourced metaseq (https://github.com/facebookresearch/metaseq) repository to improve access to research tools for training models for various NLP tasks.

Sep 2021 – Dec 2021 · 3 mos · Amherst, Massachusetts, United States

• Helped grade the papers of students taking Compsci 461/661

May 2021 – Aug 2021 · 3 mos · Natick, Massachusetts, United States

• Generated fast, efficient and accurate C++ and HDL code working on the Domain Specific Code Generation team for customers in various domains
• Created two new blocks in Simulink for pixel-wise operations on image inputs that impacts 1000+ customers using the Image Processing Toolbox.
• Built an intuitive front end in Matlab for customers to be able to specify kernel size, stride and region of interest for image processing.
• Optimized the back end to speed up C++ code generation by 20% and enabled multi dimensional input support for Neighborhood Processing blocks in Simulink

Apr 2020 – Jul 2020 · 3 mos · Kolkata, West Bengal, India

• Generated intelligible captions from large image datasets using encoder decoder architecture and soft attention to obtain benchmark BLEU score of 0.29. Represented images as vectors to obtain 2D attention map for each word in the caption.
• Obtained richer captions by generating up to twice as many regions of interest using semantic segmentation.Using Segfast allowed for more background regions of interest which were then fed into the caption generation system.
• Clustered words from multiple sentences to form dense captions and scene graphs

Jan 2019 – Jan 2020 · 1 yr · Kolkata, West Bengal, India

Dec 2018 – Jan 2019 · 1 mo · Greater Kolkata Area