Apr 2021 – Present · 4 yrs 11 mos

• ▪ Increased the new user retention for the M365 Copilot Mobile App from the baselines of 15.3% to
• 20.2% in 6 months’ time by identifying causal factors and launching product experiments with the
• PM team
• ▪ Decreased the lethal error rate for our top 100 tenants from 1.3% to 0.7% by identifying the
• problematic tenants and implementing a user level error prediction model which helped the Eng
• team design product level solutions
• ▪ Researched and implemented a SPAM/HAM detection model for the yammer team for their newly
• upgraded post feature with a precision of ~90% and recall ~93% using word2vec embeddings of
• around 300 dimensionality. Similar results with TF-IDF
• ▪ Built a supervised learning AML pipeline using MLP Neural Network to predict customer churn (1 =
• churned, 0 = retained), enabling PM teams to target users in the 0.5–0.6 churn probability bucket for
• product interventions. (F-Score was ~88%)
• ▪ Built a scalable health scoring model for mid-market tenants with sparse data, enabling proactive
• outreach and segmentation for a pilot involving 200 tenants

Apr 2020 – Mar 2021 · 11 mos

• ▪ Developed and automated a time series forecasting solution to predict Azure cloud capacity across
• regions, enabling monthly planning and resource allocation. Monitored it month over month
• ▪ Identified a high-correlation L2 metric based on core action usage to serve as a proxy for M3
• Retention%, enabling faster experimentation cycles and actionable insights for consumer app
• teams.
• ▪ Running BAU dashboards for the Yammer team with end-to-end data pipelines and supporting the feature crew with A/B testing