A few of my favorite past projects
Cloud and autoML migration
Migrated dozens of production processes from on-prem, single server SAS to AWS, elastic Spark clusters, and AutoML. Reduced costs 88% and time to market 99%. (2018) ↗
Hyper-personalization for cruise guests
Designed a real-time hyper-personalization mobile app for Carnival Cruise Line guests that collected real-time data from RFID bracelets. (2016) ↗
Personalized gift finder
Developed a personalized gift-finder for Marks & Spencer's website that peaked at a 12% conversion rate on Cyber Monday. (2014) ↗
Audience overlap
Analyzed the audience overlap of Hearst's 53 magazine websites to identify the most valuable visitors. Turned out that the biggest segment was visitors to Cosmopolitan's sex positions pages, which at the time, was difficult to monetize. (2012)
Root cause analysis of Chicagoland's cell network
Analyzed 1000 variables using AutoML to identify the cause of poor call quality. Found the same top three drivers as Six Sigma blackbelts but in 1% of the time: person years versus 4 person days. (The marketing literature says 90% reduction only because they thought that a 99% reduction would be unbelievable.) (2006) ↗
Long-tail retail
Developed new machine learning algorithms for personalization and marketing of long-tail product collections. Won the best paper award at a KXEN user conference. The algorithms are similar to what we now call hyper-parameter optimization and ensembles. (2007) ↗
Second-generation product recommendations at Amazon.com
Product manager and tech lead for Amazon.com product recommendations. I led two scrum teams: one for data and one for user interface. The main contributions were a big reduction in tech debt and increase in scalability. At the time we had 14M products and 2M daily visitors. The main innovation was asymmetric recommendations, such as people who viewed this bought that. My team generated more than $2B in incremental revenue per year. (2004)
Internet information dissemination speed
Measured how fast the phrase "Monica Lewinsky" spread across the internet in 1998. This was before Google. It took traditional media just 3 days but 3 weeks for the early web. (1998) ↗
Aquarium ecommerce metaphore
In the late 1990s it was difficult explaining to retailers how much their world was about to change so I developed a concept piece at Accenture Labs to illustrate hyper-personalization and infinite aisle space. It was a real-time, highly animated product collection browser that didn't use any keywords or taxonomies. Users simply click on items they're interested in and the products on screen change. The most similar your last few clicks, the more similar the products shown, and vice versa. It's much like what Wish.com developed 20 years later. (1999) (paper, video)
Rapide event-based programming language
Probably the most sophisticated thing I did at Stanford. The project is cited on page 100 of Dehghani's Data Mesh book. (1992)
Command and Control
First big win: Graphical command and control system developed in Ada on a Mil-Spec Data General SuperNOVA. Ada was new and we didn't have a graphics library for the SuperNOVA so I developed it. (1983)