Industry Positions & Projects
Applied Research Scientist II, Microsoft, Oct 2017 - present
Project: A Machine Learning Approach for Optimizing Azure Cloud Resource Allocation
The objective of this project is to determine the right Azure SQL DB-tier sizing per customer
with the goal of efficient resource utilization & COGS Savings.
Collected telemetry data on attributes such as DTU usage, number of active users, slow
batch queries, etc for all our customers (order of 1500).
Analyzed the current DTU usage along with all the other contributing factors and determined
the optimal tier assignment per customer using R-language.
Identified customers whom their current DB-tier assignment is not a good representative of
their usage and provided downsizing recommendations on a regular basis.
Downsizing effort is being executed in cycles. So far by downsizing our recommended candidates, we have gained a COGS savings of order of $2M per year.
By moving our customers to service fabric, current efforts are focused on clustering our customers in elastic pools in an efficient manner based on their usage history.
Project: Payment Predictor
The objective of this project is to integrate Payment Predictor into Dynamics 365 for
Finance and Operations (F&O) product. This enables an F&O user to check the likelihood
of her invoices being paid on-time while navigating through the product and to better
manage his/her cash flow.
I investigated the problem domain, relevant machine learning methodologies, and performed
predictive analytics in R-language on the synthetic data provided by Finance team. I also
worked on the featurization script in Python for the process of integrating Power AI into platform.
Project: Customers Usage Analyses
The objective was to better understand our customers' usage of F&O Entity Store & Power BI Embedded and pro-actively improved customer's experience.
I performed and bi-weekly reported exploratory and predictive analyses of telemetry data.
Research Scientist Intern, Amazon CoreML team, June-Sep 2017
Project: Informative Prior Using Empirical Bayes
The objective of my internship project was to address two existent challenges in MS and OM
applications; the absence of informative prior and inability to control parameter learning rates.
Proposed and investigated building informative prior using empirical Bayes approach in a
Generalized Linear Model.
Applied our method to a standard optimization problem, as well as an online combinatorial
optimization problem in a contextual bandit setting implemented in an Amazon production system.
Both during simulations and live experiments, our method showed marked improvements, especially in
cases of small traffic.
Our approach can be applied to any problem instance modeled as a Bayesian framework.
Product Intelligence Manager Intern, Microsoft, June-Sep 2015
Project: Predicting Customer's Business processes
The objective of this project was to predict a business process a customer is engaged in by
navigating his/her actions through Dynamics 365 for Finance and Operations product.
Investigated various machine learning methodologies to tackle the problem. Applied and
implemented Hidden Markov Model in R-language as our final solution.
Obtained over 90% accuracy on predicting customer’s behavior using D365 in F&O.
Data Scientist Intern, Microsoft, June-Sep 2014
Project: Heat Map of Search Issues in Life Cycle Services
Built a heat map of search intent and topic identification for tools provided by Life Cycle
Services (LCS) in F&O product.
Implemented and applied unsupervised topic clustering techniques to LCS search data logged
internally. Analyzed and provided preliminary classification results.
Refined and labeled topic clusters based on more scaled data. Provided Topic to business topics
index map (business process, issues, etc).