Humeil Makhija

Generated $2M revenue by optimizing Billion scale user identity graph increasing matched output id's to be exported and reducing depth and density of the same at the same time. Reduced traversal time exponentially by devising multiple Community Detection algorithms over the billion scale graph. Achieved 3-5X output exported cookies over resolved graph by traversing 4-5 levels from input matched emails id's from clients in our graph. Built a 4+ ETL pipelines using pyspark, Hadoop, Airflow and spark-ML algorithms to process Petabyte scale 3rd party data including demo-graphics, psychographics, preferences and purchase data from multiple data partners. Orchestrated Clustering, Anomaly detection and Scoring algorithms on our optimized identity graph to correctly link multiple and conflicting data elements to ensure that consumer identities can be. Built a Look-alike model based on scalable multigraph-based audience extension and scoring system improving CTR on advertisement campaigns data by 263% in comparison to other models. Applied A/B testing on ~375M user dataset over 5 online advertising campaigns increasing recall to 0.9 in predicting extended seed set users from partial seed set users in our Graph.

Automated end to end ML pipeline to optimize storage and computation cost of our Billion scale Profile Store data giving better performance affecting the segment volume export for a campaign largely as compared to most of the industry standard TTL based models. Build a robust, learnable, and incremental ML model using various Survival analysis and statistical models to purge non-active profile ID's across our system in realtime.

Relevant courses: Machine Learning, Data Science Statistics, Big Data Analytics, Algorithms and Data structures, Probability & Discrete Mathematics