Jin Y. Yi _____________________________________________________________________________ XXXXX XX XXXXX XX jyi@umich.edu XXXXXXXX, XX XXXXX XXX.XXX.XXXX _____________________________________________________________________________ SUMMARY Software engineering leader that likes to solve real-world problems for people using machine learning and amazing user experiences. Comfortable being heads-down at seed startups to leading highly visible teams in Big Tech. Generally, interested in solutions around entity association, graph theory, data mining/pattern recognition, machine learning, online experimentation and making highly available systems that solve real problems. _____________________________________________________________________________ INDUSTRY EXPERIENCE Director of Engineering Coupang Seoul, South Korea Dec 2023 - Present - Own the backend services powering pre-purchase experience in the gateway (home page), search results, and product listing pages. FTE #2 Promoted.ai Remote Feb 2021 - Jan 2023 - Built out the stream processing pipeline that fuzzy-joins user events all the way out to providing real-time count based features for our model training and delivery (Flink, Redis, Go, Python) - Architected and planned our early data infrastructure strategy focused on real-time latencies. Considered and evaluated most open and closed source data infra solutions for internet scale. - Primarily worked in the backend data infrastructure and delivery layers Googler Google Kirkland, WA Jan 2012 - Feb 2021 - Tech Lead (TL) owner of the Google Ads Budgets and Search Ads 360 (SA360) Budget Management areas - Lead SA360 Budget Management team from inception to successful launch. This was a big, multiyear effort that touched many disciplines. https://support.google.com/searchads/answer/7073010 - Extensive design and direct contributions across the full spectrum of problems touching budgets across the Ads org. Highlights include delivering solutions to algorithmic problems like automated budget allocation and budget-based bid optimization, all the way through to data visualization, exploration and simplifying the Budget UX in Ads. - Acting manager of one cross-site, and two sister teams (covering the above areas) over the course as TL - Worked extensively across the Ads org and across functional boundaries to land many successful launches and experiments - Helped lay out the integration vision between Google Ads and the Google Marketing Platform Chief Technical Officer Recess Networks, LLC Las Vegas, NV Oct 2010 - Sept 2011 - Solely designed, implemented, managed and supported the entire backend infrastructure and software to handle traffic targeting and shaping - All development was done using python 2, leveraging open source software when appropriate Computer Scientist Traffic: Automated Advertising Amazon.com Seattle, WA Sep 2009 - Sep 2010 - Worked on the automated system (Hydra) that controls the world-wide keyword portfolio on the online ad networks run by Google, Microsoft, Yahoo and Baidu for the paid traffic channel at Amazon - Lead a cross-team exploratory project that created a system that generates mappings between keywords and landing pages along with interesting conversion metrics and other statistics to be consumed as a dataset by the disparate teams across the entire traffic org. At the time of my departure, it was being consumed and tested by the SEO and Automated Advertising teams. - Developed and tested modifications to make Hydra more timely to rapid changes in bidding efficiency. Experimental results from Q4 2009 through Q1 2010 showed the potential of saving (or earning, depending on the direction of the change) tens of thousands of dollars daily. - Helped formulate the architectural changes to move from a heavily database driven backend to distributed local data files - Extended the experimental framework built in Hydra to perform automatic outlier detection - Successfully migrated (from v13) and principally maintained the Google AdWords API of Hydra starting from v2009 Computer Scientist Personalization: Similarities (Sims), Behavior-Based Search (BBS) Amazon.com Seattle, WA Jan 2005 - Sep 2009 - Researched methods to leverage item catalog data with existing customer behavior data mining techniques to better define "substitute" relationships between items - Architected the key-value lookup service that vends item similarities data to the other distributed Amazon services. The improvements over the prior architecture were (for a single machine): max throughput of 1750tps versus 700tps, tp90 latency of 1ms versus 4.5ms, and a base process memory footprint of 10MB versus 60MB. The improvements in throughput and latency allowed us to dramatically reduce our hardware costs while maintaining a very high service availability. There was also a 42% reduction in the number of lines of code for the service with a simpler client API even with a newly implemented filtering grammar. - Invented an algorithm of generating item similarities which has been surfaced on the website as "Customers Who Bought Items Like This Also Bought". In 2006, this new feature increased purchase similarities coverage by at least 6% and generated $5.5M+/yr in worldwide, incremental contribution profit. - Until mid-2007, I was the primary developer in charge of the display/ feature layer of Sims. Over that time, the percentage of worldwide retail Amazon purchases attributable to Sims features has grown from ~7% to ~11% through internationalization of our features behind a simplified code base and improvements in the user interface. Also on my watch, I managed our part of the company-wide migration to a new content rendering framework which essentially required a re-write of everything. Coupled with greatly improved documentation, these changes are allowing rapid innovation in this space for the team to build on. - Wrote many scalable backend systems that eventually replaced much of the build framework for Similarities and the rest of the Personalization department. For example, the P13nCatalogTools removed the headaches and high cost of obtaining and using the raw Amazon catalog data. - Optimized and improved the BBS build by making it extensible and applicable in domains beyond query-product relationships - Patents: - Extrapolation of behavior-based associations to behavior-deficient items US8032425 B2 - Oct 2011 - Extrapolation-based creation of associations between search queries and items US8090625 B2 - Jan 2012 - Evaluating recommendations by determining user actions and performance values pertaining to lists of recommendations US8250012 B1 - Aug 2012 - System for recommending item bundles US8285602 B1, US8290818 B1 - Oct 2012 - System for generating behavior-based associations for multiple domain-specific applications US8447747 B1 - May 2013 Software Development Engineer - Intern Customer Account Management and Payment Systems (CAMPS) Amazon.com Seattle, WA May 2004 - Aug 2004 - Owned the development of CAMPSight, a production-level intranet webtool and set of backend Perl modules, from design to live release - Restructured the Perl development hierarchy for the CAMPS (Customer Account Management and Payment Systems) group to fit newly implemented company-wide standards - Helped make design decisions on the next generation of CAM Graduate Record Examination (GRE) Instructor Graduate Management Admission Test (GMAT) Instructor Kaplan, Incorporated Ann Arbor & Kalamazoo, MI Oct 2002 - May 2004 - Trained in the Kaplan pedagogy for the GRE and GMAT tests after scoring above the percentile requirement in each test (90% GRE, 95% GMAT) - Honed teaching, presenting and mentoring skills in tutoring and classroom environments - Student feedback was consistently above the national average Junior Systems Administrator University of Michigan - LS&A System Services Team UNIX Ann Arbor, MI May 1999 - May 2001 - Compiled and ported open-source software applications on Solaris and Linux for use by all *NIX-based machines in the College of LS&A - Implemented a revised version of secure shell (ssh) that handles the proprietary security issues of the college - Developed tools and utilities to automate administrative tasks using Perl, Tcl/Tk and shell scripting Computer Consultant II University of Michigan - LS&A System Services Team NT Ann Arbor, MI Oct 1997 - Dec 1998 - Developed a centralized performance monitoring and logging system for all deployed Windows NT servers in the College of LS&A - Implemented a Zero Administration Kit lock-down of the LS&A NT image - Packaged commercial software for Windows Systems Management Server distribution _____________________________________________________________________________ RESEARCH EXPERIENCE Research Assistant Psycholinguistics Laboratory of Dr. Richard Lewis Ann Arbor, MI Oct 2003 - Dec 2004 - Explored issues of real-time computing while designing the hardware and software platform for gaze-contingent experiments - Developed an Eyelink based eye-tracking system for experimental design using Python in Windows and Mac OS/X Research Assistant Computational Linguistics And Information Retrieval (CLAIR) Group Dr. Dragomir Radev Ann Arbor, MI Oct 2003 - Jan 2004 - Developed a client-server platform to access MEAD, a text summarization platform - Extended nutch, a Java open-source web search engine, to retrieve summaries of search hits using the aforementioned client-server architecture - Investigated Support Vector Machines (SVM) for autonomous judge selection in MEAD Summer Researcher Massachusetts Institute of Technology Lincoln Laboratory Lexington, MA Jun 2001 - Aug 2001 - Researched the Dempster-Schafer theory applied in stochastic decision making and data fusion - Wrote an IDL simulation of IR sensing and classification of gray bodies to demonstrate the effect of varying a priori knowledge - Formally presented findings to researchers in the Sensor Systems group Research Assistant Psychology Laboratory of Dr. Jun Zhang Ann Arbor, MI Dec 2000 - Oct 2001 - Investigated a Bayesian-based cognitive model for a RT task - Ported a Mathematica simulation of the model to C with improvements such as reducing the algorithm's asymptotic complexity and eliminating confounds by increasing control over various independent variables - Regularly discussed topics in decision making and learning _____________________________________________________________________________ EDUCATION Master of Science University of Michigan Ann Arbor, MI Sep 2003 - Dec 2004 - MS in Computer Science and Engineering/Intelligent Systems with research interests in learning, language, cognitive modeling and classification - GPA: 7.315/8.0 (3.78/4.0) Bachelor of Science University of Michigan Ann Arbor, MI Sep 1997 - Dec 2001 - Honors Computer Science & Cognitive Science, dual concentration - University Honors, Class Honors - GPA: 3.55/4.0 (3.67 in majors) Relevant courses or transcript available upon request _____________________________________________________________________________ SKILLS/ADDITIONAL INFORMATION - U.S. Citizen - Elementary proficiency in Korean (written and oral) - Memberships: Psi Chi Honor Society, Psi Upsilon Fraternity - Languages: Python, Go, SQL, C, C++, Dart, Java, Perl, Ruby, Lisp, Erlang, Prolog - Programming Paradigms: Distributed, Parallel, Real-time, Procedural, Functional, Object-Oriented (OO), Knowledge-Based (KB), Scripting - Scripting/Markup: JavaScript, HTML, CSS, XML, LaTeX, PHP, Tcl/Tk - Network & Systems Administration: Linux, BSD, Solaris, Microsoft