Genvid pioneered MILEs—Massively Interactive Live Events—a form of new media, a hybrid of TV and games, where users decide the results of major plot points. With strong IP like the Walking Dead, Silent Hill, Borderlands, and DC Comics, my job was to architect and build a data platform—both front end data collection and backend processing—to understand our users, especially across titles.The platform, Parable, was sensitive to cost: usage patterns would vary significantly among products, each with limited lifetimes but significant audiences. As architect, I chose a system with a CDN-fronted ETL into Snowflake (a cloud superset of Postgres), bringing the amortized cost per month to tens of dollars in fallow periods; yet during active periods Parable was still below the cost of commercial systems (per MAU), yet with higher fidelity.

I founded Tiny Ventures with Sue Gardner, former Executive Director of Wikipedia, as an experimental vehicle for a number of projects. Our primary work was providing confidential tech-informed strategy consulting for clients that included private foundations and individual philanthropists, startups, large commercial tech companies, and government agencies.Our independent projects were twofold:- Freedom Fund: A technology incubator, backed by our network of private foundations and individual philanthropists, focused on bringing Y Combinator-style agile incubation to the world of public good, open-source projects. This felt particularly poignant in 2016, as the public at large became suddenly aware of the sheer scope and power that companies like Facebook had to influence elections through their data and advertising networks.- Peerless: A consultancy providing planning and conference/presentation services to technology companies to improve policy and culture for women in STEM—especially women mid-career. Most companies seemed unaware of the horrible rate of mid-career retention for high-skill women. The pitch deck gives a good overview.

I lead Wikimedia's engineering efforts to modernize its analytics infrastructure, resulting in radically improved capacity, timeliness, and reporting. This reboot touched on the full stack, from re-architecting the data processing pipeline that handles all Wikipedia traffic (nearly 100G/hour), to building a new visualization toolkit aimed at expanding self-service to non-technical users.

I architected and developed a real-time analytics system, Unicron, which powered the share-count feature of the AddThis sharing menu, at the time present on ~5%-10% of all internet domains. The service was designed to provide real-time tracking for every share of every URL (500M+ unique URLs across 20M+ domains, servicing 100k+ requests per second). As a result, it was a read-heavy workload (5% of the internet's views, at least) with sporadic-but-localized heavy write load: content that trends virally tends to get reshared, exponentially. The technical result was a unique blend of elastic data storage and the CDN: our in-cloud Cassandra cluster handled the storage, and set a TTL for requests that decayed at an inverse-frequency of the update (share) rate for a given URL (saving the service endpoints ~2-5 orders of magnitude of requests).