Anagh P. Email & Phone Number

San Francisco, California, Us

San Francisco, California, Us

All feed related data science things

San Francisco, Ca, Us

Trust and safety - Sensitive and illegal contentProvided data for TwitterSafety tweets, leadership updates to banks and regulators; highlighting wins in CSE remediation. Drove alignment across data science, legal, operations and product to launch layered remediations for text based offenses, thus increasing proactive detection rate and reducing total impressions on violating content by 65%. Identified gaps in reporting flow for CSE reports based on click data and worked with engineering to roll out fixes allowing for appropriate routing of 1000+ reports/day.Improved keyword based search suppression for known CSE terms to reduce 98% of impressions on search surface.Automated CSE suspensions for positive engagement with known CSE content leading to >400k CSE suspensions per month.Owned data science roadmap for sensitive & illegal content including setting OKRs and socializing across partner teams.Led monthly metrics review meetings for the health org and triage anomalies in metrics to ensure KRs are met.Analyzed launch of new NSFW media model which improved recall from 70% to 85% by root causing engagement losses and studying heterogeneous treatment effects across user types.Onboarded 1000+ metrics managed by airflow across multiple dimensions to an internal experimentation platform to enable measurement of causal changes and reduced analysis time from days to minutes.Created dashboards to monitor model performance for adult content classifiers and root cause issues flagged.. Researched and recommended stratified sampling for causal prevalence of health metrics and reduced variance of the relative difference estimates by ~35%, thus reducing minimum detectable effect size

Seattle, Wa, Us

Experimentation at Alexa1. Propose and set up test of percentiles for comparing tail densities of distributions for metrics in Alexa experiments. 2. Recommended metric transformations for measuring changes in high variance metrics within Alexa experiments. 3. Research methods for early detection (few hours) of severe key metric degradation in Alexa A/B tests.

Seattle, Wa, Us

1. Built customer propensity models with engagement, subscription and cross-site metrics inputs across all Prime Video locales. Scores were computed daily for the prime video customer base and used as a KPI to make product decisions in >90% experiments 2. Designed and implemented models to predict content quality based on customer engagement, content meta data- which was used to filter low quality recommended content out of homepage and reducing negative customer feedback metric by 5%3. Helped iterate faster on models by re-architecting training, validation and scoring stages from R, SQL/redshift to PySpark and reducing time to productionalize a model by ~66% 4. Worked with digital economics, finance teams to develop, implement downstream financial value metrics which helped reduce decision making time taken for engineering teams by 50% and acted as an input to optimization function for ML rankers 5. Subject Matter expert (Bar Raiser) for A/B testing specializing in experimentation design, analysis for Amazon Prime Video which conducts thousands of experiments a year. Conduct training for training new experimentation bar raisers and educating new experimenters about statistics and A/B testing 6. Analyzed experiments for personalization org and implemented 20+ metrics across different groups- streaming, subscription and financial metrics for different customer segments. Built self-serve analytical notebooks and queries to enable Product Managers to analyze experiments w.r.t. new metrics which reduced ad-hoc analysis time by 40% for the team 7. Quantified long term (annual) customer impact of experiment treatments by segregating the magnitude across different customer segments- helping the business prioritize, allocate resources to certain programs

Seattle, Wa, Us

Bentonville, Arkansas, Us

- Helped the business determine the value for Customer Initiated Discounts- Provided insights and actionable recommendations for different discounting options across various demographic cohorts- Built log linear regression, piece wise models, full models with interactions to quantify incremental basket value across different cohorts - Recommended restructuring of Promotional investments based to achieve higher engagement and revenue

Northbrook, Il, Us

- Led a team which analyzed loyalty cards performance for a retailer post launch across various geographies, customer segments, store formats and recommended changes in the loyalty card’s reward structure to reduce weekly expenditure by 40% (£2,000,000)- Analyzed change in YoY sales based on weather parameters on grocery category sales using regression models to segregate impact of external and internal factors on financial trends- Reviewed existing store layout and recommended changes based on customer buying pattern, product affinities, business constraints and summarized results using innovative data visualizing technique (Edge Bundling) in HTML and Javascript- Measured multiple campaigns (Designed and Un-designed) using a pre-post A/B testing methodology and provided improved control set enabling measurement of £200,000 weekly sales lift across various customer segments- Identified product categories facing on-shelf stocking and availability issues in stores across time and day of the week for fast moving items using variance in the volume trends with a sales potential of £800,000 per week- Analyzed customer social sentiment (using tweets, fb posts, blogs) during holiday season and identified pain points using proprietary sentiment analysis tool to compare performance with competitors and plan better for the next season