reCAPTCHA was acquired by Google in September 2009. Since then, I've been working on integrating reCAPTCHA's book-digitization technologies into the Google Books project.

As one of the co-founders of reCAPTCHA, I worked on nearly all aspects of the reCAPTCHA system. However, my main emphasis was the development of the back-end "CAPTCHA pipeline," which runs optical-character recognition (OCR) software on scanned documents, creates CAPTCHAs to help digitize these documents, distributes these CAPTCHAs to front-end servers, and uses a probabilistic language model to analyze the OCR guesses and human answers to produce high-quality digitized output.

Thesis: Thresholded-Rewards Decision Problems: Acting Effectively in Timed Domains.

For my doctoral thesis research, I investigated how teams of agents or robots can act near-optimally in timed, zero-sum games. In these domains, teams need to act in order to maximize the probability of winning, not necessarily to maximize their score. The general idea is that a team which is losing should act more aggressively to try to even the score, and a team which is winning should act more defensively to try to preserve their lead. I presented algorithms which compute optimal policies for these timed domains, using Markov and semi-Markov models (MDPs and SMDPs). I also analyzed how a team should change strategy in response to an opponent whose behavior is initially unknown but slowly reveals itself during execution. My thesis work was applied to the RoboCup robot soccer competition, a simulated Capture the Flag domain, and book digitization.

In addition to my studies, I was an undergraduate research assistant on three research projects: multi-robot systems, the Trading Agent Competition's supply-chain management game, and intelligent transportation systems.