Apr 2023 – Present · 3 yrs 2 mos · Seattle, Washington, United States

Aug 2022 – Dec 2022 · 4 mos · Pittsburgh, Pennsylvania, United States

• 10-601 (Masters): Introduction to Machine Learning

Jan 2022 – Aug 2022 · 7 mos · Pittsburgh, Pennsylvania, United States

• 11-785 (PhD): Introduction to Deep Learning

May 2022 – Aug 2022 · 3 mos · Seattle, Washington, United States

• 1. PyTorch4All: Migrated the tech stack of ML Search modules (low-level and high-level) from Caffe2 to PyTorch. Ensured consistent model performance across frameworks through unit testing for latency and accuracy. Also implemented Torch Scripting wrappers to optimize the model run-times.
• 2. Position Debiasing in People You May Know (PYMK) product: Improving recommendations (friending score) by feature engineering in multi-task multi-label (MTML) models

Apr 2021 – Aug 2021 · 4 mos

• Led the design and development of an AI-powered question generation (MCQs, True/False etc.) and question answering system, leveraging a fine-tuned combination of BERT, T5 and GPT-2 state-of-the-art transformer models. Deployed as serverless microservice architecture using Docker and AWS Lambda.
• Edit and modifications were allowed to the generated worksheet before exporting it as plain text or PDF.

Jun 2020 – May 2022 · 1 yr 11 mos · Remote

Mar 2019 – Aug 2020 · 1 yr 5 mos

• Trained an LSTM-CRF Named Entity Recognition (NER) model to extract the project name, locality, property type, and price from real-estate classifieds, thus automating the inventory (supply) creation on the platform.
• Developed a pipeline of different CNN models to detect and discard faulty images containing, competitor logos, text, and NSFW content. Accepted images were then classified into different room categories like bedroom, bathroom, kitchen, etc. This reduced the time-to-go-live of a new listing from 3 days to 1 day.

Nov 2018 – May 2021 · 2 yrs 6 mos

• Developed a paid course "The Android App Market on Google Play" on DataCamp
• 57,000+ learners
• Mandatory module in "Data Scientist with Python Career Track" specialization.
• Created "Guided" and "Unguided" versions of the course aimed at instructing learners to clean, visualize and analyze Google Play Store apps. Also included is an exercise on sentiment analysis of the user reviews.

Jul 2018 – Sep 2018 · 2 mos · Greater New York City Area

• Web scraped 10k+ Google Play Store apps using Selenium to investigate the Android market. Incorporated basic sentiment anaylsis on user reviews. Currently, the dataset ranks as third-highest upvoted dataset on Kaggle.
• Built an interactive R Shiny search engine for comprehensive analysis of 1.5 million traditional dishes from various world cuisines, allowing users to explore characteristic culinary patterns by regions and ingredients.
• Developed a Flask web app to recommend similar-styled clothes based on eight unique styles modeled using LDA from over 5k web-scraped product descriptions. Found similarity between images using pre-trained CNN VGG16 model and cosine similarity.
• Modelled Hong Kong Jockey Club horse races and estimated the win probability to place value bets and maximize ROI. Used linear regression and decision tree along with Kelly Criterion for simulating betting procedures.
• Evaluated machine learning models (Lasso, Ridge, Elastic Net, Random Forest, XGBoost) for housing price prediction (closed Kaggle competition). Optimized performance by ensembling and stacking techniques.

Jun 2016 – Mar 2019 · 2 yrs 9 mos · Pune, India

• Automated Interpretation of Organization Charts: Used OpenCV for box detection alongwith OCR and graph algorithms to find hierarchical connections between entities. Eliminated manual QA checks by ~55%
• Designed and developed an end-to-end system for real-time monitoring and analysis of call-centre voice calls using Google Cloud services. Expedited customer grievances redressal process by taking proactive measures.

Jan 2016 – May 2016 · 4 mos

• Conducting programming labs
• Designing lab assignments and evaluating student performance

Jul 2015 – Dec 2015 · 5 mos

Jan 2015 – Jan 2015 · 0 mo · India

• Track: SENSORS ACROSS SCALES - Solving Social Problems with Technology
• The project, Colored Yarn, at the MIT Media Lab India Design Innovation Workshop 2015 has been created to provide technology advantage to the weaver community of India. It is a prototype that can color the yarn on the fly!

Dec 2014 – Jan 2015 · 1 mo

• This project under the Carnegie Mellon University Internship Program, 2014, entitled 'Predictive Policing using Regressive Time-Series Analysis', focused at mining crime reports from the Web, cleaning the data and encoding various statistical models for crime-rate prediction.
• The project culminated into a research paper accepted at WOCN 2015, an IEEE conference.
• Currently being used as a pilot project in different police stations in India.