• Led a 9-person remote team of senior- and lead-level machine learning engineers• Improved team efficiency by hiring dedicated MLOps engineers; functionally managed a product analyst team to support testing and adoption of machine learning features• Supported the migration of Skillz advertising spend onto Aarki’s DSP; took 35% of overall DSP spend in-house, up from 8% at the time of acquisition• Established a process for hypothesis-based machine learning feature development, with intermediate result review and go/no-go checkpoints

• Developed machine learning engine for impression value estimation and bid optimization in real-time ad auctions, generating over 2.5B bids per day• Improved user response prediction models by engineering device-level features; resulting model lowered campaign CPIs by 50%• Developed a first-price auction strategy and bid optimization layer; by 2020, 90% of auctions were first-price• Developed and deployed an end-to-end machine learning pipeline on-premises; deployed models to over 1,000 bidderservers across five data centers, ensuring auditable predictions and sub-10ms latencies• Integrated model building into campaign management UI; saved 2 hours of DS/analytics time per day, and reducedcampaign launch time by 3 days on average• Developed models for buying limited ad-tracking traffic ahead of Apple’s ATT release; 90% of iOS campaigns werebidding on SKAdNetwork-enabled traffic within 90 days of ATT release• Designed an algorithm for generating optimal SKAdNetwork campaigns to maximize information gain for downstreamlearning and analytics• Provided client-facing support for key accounts totaling over $2M in monthly ad spend on Aarki

• Designed and executed a ground-up rewrite of Aarki’s in-house creative studio, a web-based editor for designing interactive rich-media HTML5 creatives, using state-of-the-art Javascript MVC frameworks and web APIs; added responsive ad capabilities and cross-platform animation support• Added support for constrained multivariate creative; enabled 3x ROI improvement for re-engagement campaigns through user-level personalization of ad creatives• Developed a statistical method for detecting and filtering accidental ad clicks based on coordinate click logs; implemented heatmap visualizations of creative clicks enabling quick detection of creative rendering issues• Designed and implemented novel data structures and algorithms for storing, updating, and querying histograms of two-dimensional data