Working as tech lead to develop ML models that provide next best action recommendations for tenants based on their usage of Microsoft 365.

Working collaboratively across product, marketing, engineering to deliver ML driven solutions to complex business problems.

Worked extensively with M365 product and engineering. Built data-driven products to enable smarter recommendations for admins and end-users to ensure the best usage of M365 services. (churn, upsell, next best action recommendations, propensity models).

Preprocessed large amount of M365 usage telemetry using Spark with Azure ML environment to enable opportunities / recommendations for Microsoft customers and Partners. Model outputs power internal dashboards for internal Field and business teams.

Building NLP machine learning models for the Healthcare space using Azure ML and Databricks

Hospital Avoidable Admission Prediction

1. Preprocess structured data of member’s medical history using Spark dataframes and Databricks within the Azure Machine Learning environment.

2. Productionized a predictive model that predicts the probability of a member having a hospital admission in the next 60 days that could have been avoided with proper intervention from a care manager.

3. Developed a Doc2Vec (NLP) recommendation engine that matches member clinical profile (as a document) to relevant and actionable interventions.

4. Design hypothesis test to measure 90-day avoidable admits after randomization of observations into control and test groups.

5. Developed Microsoft Azure Data Factory pipelines to deploy models and enabled flow to display model outputs as recommendations to practitioners.

Personalized Product Recommendations for B2B Customers

1. Developed a collaborative filtering model with SVD matrix factorization to provide personalized product recommendation to B2B customers at Wayfair.

2. AB test results show that implementation led to a 2.07% lift in conversion and brings in $2million incremental yearly revenue for Wayfair.

3. Productionized a total of five collaborative filtering models using SVD for different Geos and specific use cases. All of these models bring in an incremental yearly revenue of $4million to Wayfair.

4. Python for development. Developed Airflow DAGs for putting all models into production

Customer Product Affinity Score v1.0

1. Conceptualized and productionized a random forest classifier that scores a customer’s affinity to a product on Wayfair’s website using features like reviews read, number of visits to the product page, etc.

2. This model is used in retargeting ads at Wayfair by showing the product with the highest affinity score to the customer. The model has yearly incremental revenue impact of $10million.