Jun 2021 – Oct 2022 · 1 yr 4 mos · Kolkata, West Bengal, India · Remote

• First ML Engineer hired at Carted :-)
• Developed scalable data preprocessing modules that can operate with raw and unclean HTML data. Here's a relevant blog post that discusses some parts of it: https://bit.ly/dataflow-pipelines.
• Developed modules for training attribute extraction models. These models can extract global attributes (like price, title, description, etc.) from product webpages. This reduced the need for writing custom website wrappers.
• Set up the MLOps tooling for running our preprocessing and training code at scale on Google Cloud Platform. We leveraged various tools for doing this, e.g., Dataflow, Vertex AI, BigQuery, etc.
• Developed utilities (data preprocessing, training, and evaluation) for pre-training on an internal large text corpus. Pre-training strategies include masked language modeling (RoBERTa), and latent representation prediction (Data2Vec).
• Worked on improving the latency of our product categorization service from 61.63 ms to 4.63 ms! This blog post discusses details: https://bit.ly/efficient-ml-api.
• Worked on improving the hardware provisioning workflow for running our ML experiments. This helped us eliminate the redundant steps and achieve reproducible hardware provisioning. Here's a relevant blog post: https://bit.ly/terraform-ml.

Oct 2019 – Oct 2020 · 1 yr · Kolkata, West Bengal, India · Remote

• Weights and Biases (W&B) Authors Program (https://www.wandb.com/authors) is an initiative run by the company to put together the best Machine Learning (ML) and Deep Learning (DL) practices in the form of articles and reports. As a Weights and Biases Author, I am responsible for the following:
• Create technical articles and reports that showcase a project or concept relevant to the field of ML and DL.
• Test out new product features and provide constructive feedback.
• Suggest new product features.
• All my W&B articles and reports are attached below.

Jun 2019 – Jun 2021 · 2 yrs · Kolkata, West Bengal, India · Remote

• I wore multiple hats in my role. From researching novel ideas for blog posts to implementing them, I helped PyImageSearch build a strong suite of Computer Vision and Deep Learning projects. The tasks involved in these projects include image generation, data input pipeline acceleration, training pipeline optimization, spatio-temporal modeling, etc. These projects helped PyImageSearch acquire new customers thereby increasing its profits. Here are some of the key responsibilities I performed:
• Acted as the key developer and researcher for a CVPR 2021 competition and led our team to the top 10. The competition report is attached below. To this end, our small team developed a simple approach to perform knowledge distillation for compressing larger models into smaller ones.
• Developed TPU-compatible training pipelines thereby reducing the overall costs by ~30%.
• Responsible for developing the majority of the code for the chapters of the Complete Bundle of our book Raspberry Pi for Computer Vision (https://pyimagesearch.com/raspberry-pi-for-computer-vision). These chapters are themed at the intersection between Computer Vision and Deep Learning.
• Co-developed a project on image colorization that achieves near state-of-the-art performance.
• Developed and took ownership of our Jetson Nano .img (http://pyimg.co/ozukh), RPi4CV AWS AMI (rpi4cv-pyimagesearch-v2.2 [Zone: Oregon]), and GCP environments.
• Developed a utility to automate our zip file uploads to AWS S3 Buckets reducing manual efforts by ~30%.
• Provided Q&A support to PyImageSearch readers.

Mar 2019 – Dec 2019 · 9 mos · Kolkata, West Bengal, India · Remote

• FloydHub (https://www.floydhub.com/) is the Heroku for Deep Learning enabling the users to focus on the science and taking care of the infrastructure, environments, deployments and version controls. As an AI Writer at FloydHub, my responsibility was to write on crucial topics centered around fields like Data Science and Machine Learning (sometimes Reinforcement Learning too).
• FloydHub articles are aimed to provide the readers with the right amount of theoretical intuition along with practical implementations. One can find all the FloydHub articles here: https://blog.floydhub.com.
• My articles are attached below. Feel free to let me know your feedback.

Aug 2018 – May 2019 · 9 mos · Kolkata, West Bengal, India · Remote

• DataCamp helps companies and individual learners answer their most challenging questions by making better use of data. Our learners build and maintain data fluency on the world’s most advanced online learning platform for data science and analytics.
• Responsibilities:
• Developed the following projects - Predicting Credit Card Approvals and Analyze International Debt Statistics. Both have a median rating of 4.5.
• Developed the following practice pool - Advanced Deep Learning with Keras in Python. A demo is available here - https://www.loom.com/share/63f758542f764884a0c70547e2fa0e47.
• Wrote technical tutorials based on Machine Learning for DataCamp Community. Following are the two best ones having more than 50K views - Turning Machine Learning Models into APIs in Python, Simplifying Sentiment Analysis in Python.
• Apart from these, I handled a few of the DataCamp's outreach activities as well.

Jan 2018 – Aug 2018 · 7 mos · Pune Area, India · On-site

• Part of TCS Research and Innovation. TRDDC is also the first corporate research lab of India.
• Domain of work: Data Privacy
• Product: Crystal Ball Consent Management Solution (https://on.tcs.com/2RR9sVe)
• Executions:
• Implemented web services covering data privacy practices for GDPR enforcement.
• Developed a demo UI in order to demonstrate the web services to both internal and external clients.
• Implemented test cases and eventually increased the code coverage of existing services to 89%.
• Prepared Swagger API documentation for facilitating interactivity to API documentation.
• Developed Postman scripts to automate API testing workflows.
• Automated the task of listing out web services with their endpoint URL from code.
• Implemented a utility with Spring Batch for managing DB transactions in an efficient manner.
• Involved in thorough functional and security testing of the web services.

Oct 2017 – Jan 2018 · 3 mos · On-site

• Responsibilities:
• Fixed service defects both at production and hotfix level.
• Designed a system for ensuring the service responses' structure according to a given specification.
• Modified existing test cases in order to facilitate more method scenarios and eventually increased code coverage in all three levels: Service level, Business Logic level, and DAO level.
• Prepared Swagger documentation of the web services.
• Client:
• Information major, USA.

Jul 2017 – Oct 2017 · 3 mos · On-site

• ILP (Initial Learning Program) training on Java and related technologies.
• Projects:
• 1. Retail Banking Solution (Java EE, Oracle 11g XE, Apache Tomcat, HTML-CSS-MaterializeCSS, JavaScript, jQuery, JSP):
• The project was a part of the training. Designed and developed a solution for in-house banking activities like profile creation, account summary generation etc on both front-end and back-end levels.
• 2. IoT Project using Ultrasonic Sensor and TCUP (Core Java):
• This project was an addendum to the training period in TCS, taking place in the ILP Innovations Lab. We used ultrasonic sensors, published the data to the cloud and pulled them back to perform tasks based on the data obtained.

Dec 2016 – Feb 2017 · 2 mos · Kolkata, India · Remote

• I was responsible for scrapping specific data of schools (of West Bengal) from their websites and gather them in a certain format. I used BeautifulSoup as the main Python library for this purpose.