Cristos J. Goodrow

YouTube (a subsidiary of Google) Vice President, Engineering
March 2011 to present I lead the 700-person engineering team for the Core Experiences of YouTube, including development of the main YouTube app across mobile and desktop, as well as the search and recommendations systems. YouTube is among the largest search engines in the world, and the recommendation system is even larger and has won an Emmy Award. Many of these systems use very large scale machine learning, especially deep neural networks using reinforcement learning. I led changes to the optimization objectives from views to watchtime, and ultimately towards long-term viewer satisfaction. I helped grow YouTube from 100M hours of viewership per day in 2012 to over 1B hours per day, used by billions of people around the world. I also work to ensure that YouTube grows responsibly, reducing the prevalence of harmful misinformation and hate, and promoting authoritative sources of information, while ensuring that the systems remain fair by developing advances in Machine Learning Fairness.
Google Director of Engineering
February 2008 to March 2011 Responsible for Google Shopping (formerly Google Product Search), which is available in several countries.
A9 (a subsidiary of Director of Search Engine Software
July 2005 to January 2008 As Director of Search Engine Software, responsible for developing the software that performs search on,,, and many more of the largest eCommerce internet sites. Directed the Search Relevance, Search Infrastructure, Search Analytics, and Search QA groups, hiring top-level Directors and Managers to lead the groups. Improved the quality of search results on Amazon and other sites through increased use of customer behaviour and probabilistic techniques. Dramatically increased the scalability and maintainability of the real-time updatable search system. Initiated consistent and accurate measurement of the quality of search results over time. Member of the Executive Management Team, reported to the President of
DemandTec Director of Engineering
November 2001 to March 2005 DemandTec provides demand-based price, promotion, and markdown optimization software-as-a-service to leading retailers, including Best Buy, Safeway, RadioShack, WalMart, Monoprix, and many others. As Director of Engineering, I was responsible for directing development groups concerned with the most critical areas of DemandTec's offering including the database group, the econometric modelling group, the optimization group, and the OpenLink data integration group. Led the entire development organization to successfully deliver DT 4.0 in May 2004. During development of DT4.0, I coordinated all development and testing activities, measured progress and quality, and reported directly to the CEO.
The econometric modelling group and the optimization group are together responsible for implementing all scientific algorithms used in the DemandTec system. The modelling group invents and implements algorithms that create a statistical model of consumer demand from historical sales information. Techniques used include mixed-effects regression (Bayesian shrinkage), Affinity Relationship Mining (ARM), non-linear optimization, and clustering. The optimization group uses non-linear programming to optimize merchandising decisions (prices, promotions, markdowns) for increased revenue, profit, or sales volume, subject to a hierarchy of flexible constraints or rules. The code is implemented using the General Algebraic Modeling System (GAMS) and must efficiently select prices or promotions for thousands of products in hundreds of stores.
The database group is responsible for the data management strategy, database schemas, database deployment and data migration for all components of the DemandTec system. The OpenLink team is responsible for integration to other systems, particularly for moving data between systems and through the DemandTec system. Both the database and OpenLink teams have been challenged during my tenure with customer data sizes that have gone from several Gigabytes to several hundred Gigabytes, and with the number of customer databases increasing from three to over sixty. These teams have met those challenges by switching from SQLServer to DB2, by becoming accomplished experts in DB2 development, administration, performance, scalability, and by implementing novel dataflow techniques for processing large amounts of data.
TeaLeaf Technology Director of Engineering
March 2000 to November 2001 As Director of Engineering, responsible for managing several groups of engineers and leading them to deliver software releases, from providing vision and scope for a release, through requirements, design, development, system test, delivery and support. The combined size of these groups ranged from four to thirteen as TeaLeaf increased and reduced force, and they included the Quality Assurance group as well as development groups. Additionally, I was responsible for designing and implementing improvements to the development infrastructure and methods, including evaluating and selecting development tools and practices.
My development groups at TeaLeaf were concerned mainly with organizing, storing, and analyzing behavioural data captured from a web site by the TeaLeaf TeaCommerce Platform. This captured data is imported ultimately into either a SQL Server or Oracle data base, and may pass through several other formats. Once in a data base, the data is available for on-line analysis by TeaLeaf's OLAP discovery product OpportuniTea, and for aggregate analysis by TeaLeaf's ClariTea reports product.
Impresse Corp. Manager, Senior Software Architect
August 1998 to March 2000 Conceived and defined, with product marketing, collaborative project tracking features for the business-to-business eCommerce internet service. Led and managed a team of five engineers to design, implement, and deliver those features. Implementation included user interface, middleware, and data base design, and involved writing Java Server Pages (JSP), Extensible Markup Language (XML), and Java Servlets to run within the BEA Weblogic application server.
Led and managed a team of four engineers to extend and maintain the planning and scheduling system, as well as the overall architecture. Worked with product marketing and visited customer sites in order to identify areas in which to apply our technology. In particular, we focused on extending the planning system using constraint propagation and a variation of partial-order planning, enabling users to create and estimate manufacturing plans for printed materials.
Responsible for design and implementation of a system to allow for planning, scheduling, and execution of tasks in an asset-intensive environment. The architecture incorporates real-time information from the execution engine, provides for investigation of "what-if" scenarios and automatic re-scheduling, and can recover the shop-floor state for restart in case of failure. The system is written in Java, implemented as Enterprise Java Beans (EJB's) and Remote Method Invocation (RMI) Servers within the BEA Weblogic application server, and the initial platform is Windows NT.
Peoplesoft, Inc. Mathematician
Summer 1997 Designed and developed a system for representing and solving manufacturing scheduling problems using linear programming. This system, which began as the prototype I had built the previous summer, was written in C++ and has been incorporated into the overall product.
Summer 1996 Investigated methods of using linear programming to solve scheduling problems from manufacturing. Developed a prototype in C++ to restrict the general scheduling problem to a network flow problem, which was solved using the CPLEX optimizer.
Andersen Consulting Senior Consultant, Analyst
Summer 1995 Created the overall error-handling approach for the Customer Care component within the billing system of a large telecommunications company; designed and implemented the solution within the architecture of a CASE system developed in C++ by Andersen Consulting and running on a UNIX system.
September 1991 to August 1994 Led and managed a team of five engineers through design and programming of the bill printing and archiving architecture within a billing system for a large utility in Northern California. The billing system uses the Sybase RDBMS and runs on several Sun UNIX servers networked to Windows NT clients.
Designed, planned, and led the execution of a performance test of the package tracking system for a large express mail carrier. The system was developed in C and UNIX and runs on six HP servers networked (via X.25) to service centers throughout the world.
Designed, planned, and supervised the system test of the technical architecture for an inventory management system at a large telecommunications company. Managed three engineers during integration and system test. Testing included point-of-failure analysis and statistically sampling the input/failure space of the communication layer.
Designed and built the client/server architecture for communication between hand-held DOS bar-code scanners and an HP UNIX host in the automatic inventory management system for a large telecommunications company. Implemented the client-server communication protocol, including Z-modem and DES encryption.
Designed and built synchronous and asynchronous interfaces between Sybase and DB2 data bases across multiple UNIX platforms for a trade authorization system at a major asset management company.
Supported and analyzed the billing system at a large telecommunications company, optimizing the processing cycle and tuning the automated job scheduler to reduce wrap-around time and increase throughput. Proved that this particular scheduling problem is NP-Complete and tried various techniques to find approximate solutions in polynomial time.
Patents and Inventions
Cross media type recommendations for media items based on identified entities (US 8856111), jointly, assigned to Google Inc
Learning characteristics for extraction of information from web pages (US 9443250), jointly, assigned to Google Inc
Adaptive recommendations of user-generated mediasets (US 9659093), jointly, assigned to Google Inc
Method and apparatus for providing price updates (US 6851604), jointly, assigned to DemandTec Inc
Dynamic creation of workflows from deterministic models of real world processes (AU 6432599, EP 1135737, US 6279009, WO 0034909), jointly, assigned to Impresse Corporation.
Methods for creating aggregate plans useful in manufacturing environments (AU 1317100, US 6278901, WO 0038092), jointly, assigned to Impresse Corporation.
Method and apparatus for order promising (US 6321133), jointly, assigned to Impresse Corporation.
Method and apparatus for predicting the possibility of additional orders (AU 1116500, EP 1135738, WO 0034908), jointly, assigned to Impresse Corporation.
Method and apparatus for creating adaptive workflows (AU 1315700, US 6546364 WO 0038091), jointly, assigned to Impresse Corporation.
Technical background
Computer languages: Java, C++, SQL, C, SAS, GAMS, Lisp, Assembly (M68000), and Prolog
Data Bases: DB2, Oracle, MySQL, SQLServer, and Sybase
Operating systems: UNIX (Linux, AIX, Solaris, HP-UX), Windows
Currently on leave-of-absence from Ph.D. program in Mathematical Logic and Theoretical Computer Science, Department of Mathematics, U.C. Berkeley
Master's Degree in Mathematics, May 1996, University of California, Santa Barbara
B.A. in Mathematics and Computer Science, May 1991, College of Arts and Sciences, Cornell University
Activities and honors
Led the YouTube team that won a technical Emmy Award for Personalized Video Recommendations.
San Mateo Public Schools Volunteer at Burlingame High School, aiding in mathematics and computer science.
1996 Raymond L. Wilder Award, honoring outstanding first-year graduate students, from the Department of Mathematics, U.C. Santa Barbara.
San Francisco Public Schools Volunteer at Galileo High School, aiding in mathematics and computer science.
Co-chair, Cornell Alternative Spring Break organization, concerned with organizing and running community service trips during spring break.
Quill & Dagger senior honor society (Cornell University).