Carl Livadas, Ph.D.Sojern
VP of Engineering and Data Science
255 California St, Suite 1000
San Francisco, CA 94111
|Resume: [pdf,ps] CV: [pdf,ps,txt]|
Company Profile (circa 2017):
Specializing in traveler path-to-purchase data for over a decade, Sojern is travel's direct demand engine for thousands of brands --- from global enterprises to boutique operators --- across the hotel, airline, cruise, transportation, tourism industries and more. Analyzing the world's travel intent signals with its proprietary data science methods, the company delivered $7B in direct bookings for its clients to date by activating multi-channel branding and performance solutions on the Sojern Traveler Platform. Recognized on the Top Company Cultures list by Entrepreneur Magazine, Sojern is headquartered in San Francisco, with teams based in Dubai, Dublin, Hong Kong, London, Mexico City, New York, Omaha, Paris, Singapore, and Sydney.
Company Profile (circa 2017):At the forefront of the in-house advertising movement, Nanigans arms marketing teams with the best software to manage their digital advertising in-house. For nearly five years, Nanigans has developed the industry-leading advertising automation platform to empower CMOs to achieve their social and mobile performance marketing goals at scale. Nanigans is a leader among ecommerce, gaming, and other pure-play internet companies, enabling in-house teams to take control of their advertising through sophisticated workflow automation, predictive optimization, deep data integrations, and real-time and lifetime reporting tools.
Role:As the VP of Optimization (RTB), I led the engineering and data science team responsible for building NanML, Nanigans' RTB machine learning pipeline and bid evaluation tier. NanML included: 1) automated DAG-based ETLs to prepare the training data sets, 2) automated ML model training including feature generation, feature filtering, training, calibration, and evaluation steps, 3) hyper-parameter exploration on a per-model basis, 4) automated model deployment, and 5) a horizontally scalable bid evaluation tier to evaluate each RTB bid request on the behalf of each ad (post targeting matching). NanML supported rapid iteration by allowing Nanigans to explore new features, build new models, and experiment with new approaches to bidding within minutes, all without the need to deploy new code.
Company Description (circa 2013):
inPowered's mission is to make people better informed and more influential. Every day, the company helps thousands of people make informed decisions by connecting them with content written by trusted experts. Fortune 500 brands and SMBs use inPowered's platform to discover and promote credible expert opinions to generate new customers. inPowered is a privately held company headquartered in San Francisco with offices in New York and Toronto. Visit inpwrd.com for more information.
Built out and led inPowered's data platform engineering team and led all of inPowered's data science efforts. Led the design and managed the implementation of inPowered's 2nd generation data platform infrastructure to handle 100x+ the data volumes and afford near real-time data updates. Spearheaded the transition to a Java-based infrastructure to ensure performance, maintainability, and extensibility. Championed the adoption of new technologies/practices as appropriate, leveraging the learnings of the broader engineering community to keep inPowered at the technology forefront, to ensure scalability and performance, and to afford sharing of engineering resources. Moderated the adoption of novel services/technologies (AWS and 3rd party service providers) to ensure the right cost/benefit trade-offs for inPowered. Instigated the use of a Hadoop/Hive based infrastructure to mine audit logs for infrastructure and data KPI reporting. Led the algorithm design of: 1) article classification, 2) article sentiment analysis, 3) article ranking, 4) estimation of article readership based on the article's social media engagement footprint, and 5) ranking of authors within topics of interest.
Led the engineering efforts of KAYAK's Sunnyvale office focusing on KAYAK's email, deals, online advertising, and parts of the mobile application products. As a principal scientist, focused on efforts to regionalize/personalize KAYAK's products and optimize performance.
Contributing to several aspects of the Distributed Detection and Inference (DDI) project. DDI is a collaborative worm detection system involving local and global detectors on end-hosts. Local detectors issue and disseminate local infection reports indicating whether the end-host is infected. Global detectors collect local infection reports and issue system-wide alarms when the infection evidence has sufficiently been corroborated. Prior and current work includes the design and implementation of: 1) an adaptive local detector that adjusts the threshold of issuing alarms based on a learned model of its behavior, 2) a faithful analytic model of the behavior of DDI; this model is critical in understanding the behavior of the system, evaluating its performance and scalability properties, and exploring its parameter space, and 3) efficient and scalable gossip-based messaging and membership services for DDI.
Conducted cutting edge research in computer networks. Contributed to the following projects: ZombieStones; A system that leverages machine learning techniques to identify network connections that are part of suspicious botnets. IP-SPOOR; An entropy-based study of how to place network traffic monitors for effective IP packet traceback (i.e., tracing an IP packet involved in a cyber-attack back to its true source host). Stepping Stones; A system that identifies interactive connections that are used in sequence to obfuscate the origin of a cyber-attack. Stingray; An insider threat detection system that uses Bloom filters to efficiently log large amounts of network traffic and principal components analysis and machine learning techniques to detect network traffic anomalies. Performance Evaluation of a Proprietary Network; An evaluation of the performance of a proprietary network (disclosure of details of this project is restricted).
Designed, modeled, and analyzed (both formally and through simulation) a variant of the Scalable Reliable Multicast (SRM) protocol that exploits packet loss locality through caching.
Prosopa.com leveraged audio and video technology developed at the MIT AI Lab to deliver photo-realistic talking faces driven by text or audio. Prosopa.com's target markets included personalized video advertisements, customer support applications, and automated news-readers.
Researched the area of persistent TCP connections, a feature included in HTTP 1.1. Developed a simulator to evaluate the performance of persistent TCP connections between proxy and back-end servers.
FairTrust, Inc. delivered consumer-to-consumer trustworthiness rating services to online communities such as E-bay, Inc. FairTrust, Inc. has since transformed itself into OpenRatings, Inc. (www.openratings.com).
Participated in research involving the modeling and verification of hybrid systems using Hybrid I/O Automata. Performed modeling and verification of Raytheon Corporation's Personal Rapid Transit system (PRT 2000TM) and the Traffic Alert and Collision Avoidance System (TCAS) of commercial aircraft.
Math for Computer Science (6.042), Spring 2000
Lab. in Software Engineering (6.170), Spring 1997
Lab. in Software Engineering (6.170), Spring 1996
Computer Language Eng. (6.035), Fall 1995.
Developed a formulation of H2 robustness criteria for systems involving real parametric uncertainties in terms of linear matrix inequalities (LMIs) and an iterative robust H2 controller synthesis scheme.
Last modified: 2010/05/31, 23:32:00.