Concurrently with my time at Cortico, and part-time at various levels since 2020, I've been a staff member at the Lab for Social Machines at the Media Lab. Here I do a combination of software engineering and research: I write software used by students to acquire and analyze media data (social media, online news articles, and talk radio), help with courses, and collaborate with grad and undergrad students on their research projects and occasional papers. My areas of focus have been conversation analysis; topic modeling in search and discovery interfaces; fairness in automatic speech recognition systems; polarization and misinformation in social media; and applying NLP models to social science data sets. [Python, JS (React); AWS, GCP; PyTorch]

Datamuse is a venture I founded in 2000 to make reference and brainstorming tools: websites, mobile apps, and developer APIs related to words. While Datamuse has been a side project to my "day jobs" for twenty years, I spent the 18-month period after Sift focused on it full time. My primary areas of focus were updating the OneLook Reverse Dictionary tool; building the Datamuse API and working with educational app developers to integrate it; adding a Spanish-language rhyming dictionary and thesaurus; and adding an advanced search feature to RhymeZone. More recently I've focused on machine-learning approaches to finding sentences that are helpful for vocabulary-building, and semantic search tools for finding lyrics and quotations. Since 2015 I have hired and supervised the work of more than 15 contractors and co-op students who have helped with these projects. [Python, Java, Javascript; AWS, TensorFlow, Spark]

Cortico is a non-profit spinoff of a research group at the Media Lab focused on deploying tools that make sense of and improve upon conversations -- both the kind that happen online, and the kind that happen in person. Early on I implemented a Twitter monitoring tool specialized to civic issues which we tested with two local news organizations. I then built (and continue to maintain) a transcription pipeline and search engine for U.S. talk radio that has since been used by journalists, social science researchers, and media studies students. Later, as part of a larger engineering team, I helped create Earshot, a now-defunct public sphere search engine, and Local Voices Network, a system for holding facilitated group discussions in communities around the U.S. My role on LVN was to write the speech processing pipeline and conversation analysis engine that feeds the web application that journalists and community members use. (I continue to spend some time with Cortico as an advisor while I have shifted to a research role within the lab that spawned it.) [Python, JS (React), some C++; AWS, GCP]

At Sift I helped build a fraud detection API and review tool for websites and mobile apps. I wrote the first versions of our data visualization console and feature extraction framework, and later became the tech lead for the machine learning team as the company grew. As the first non-founding employee I was involved in most aspects of the company's growth, including recruiting, customer acquisition and support.

Ads stats / ads quality team (2004 - 2008)- Wrote data analysis tools and designed analyses to support the early growth of the AdWords and AdSense user/advertiser/publisher ecosystem.- Started and led several projects that grew into small teams, each related to ranking/pricing ads or analyzing AdWords performance. Key projects were: "Smart Pricing"​ (conversion-adjusted click pricing); expanded broad match (keyword synonym expansion); the live A/B experiment analysis dashboard used throughout the ads and search teams; and an IP geolocation system that enhanced the performance of local-targeted AdWords ads.Speech team (2008 - 2010)- Wrote metrics dashboard and logs analysis tools supporting mobile voice recognition efforts- Wrote pipeline to build training sets of voice search logs based on user confirmation measures- Co-wrote infrastructure for user-level adaptation of voice search acoustic modelsEconomics team (2011)- Developed an internal tool used by support teams across the company to identify trending product questions in Google search query logs- Prototyped an interactive visualization of correlations between topics in the search query logs20% Projects and misc- Implemented an API for Google Trends. (It didn't launch)- Worked with Google.org on search query analysis, building two prototypes that were the precursors to the Google Flu Trends and Google Correlate products- Co-authored 5 patent applications; 3 granted so far- 2 EMG awards[C++, Python, Java]

Software engineer on the Lycos Research team focused on analyzing search query logs to make improvements to the Lycos and HotBot search engines and to increase revenue for the company. Projects included query recommendation ("related searches"), language identification, industry vertical clustering, and ad targeting tools.

3-time intern with the Speech Technology Group at Microsoft Research (Jan-Sept '94, summer of '95; summer of '97). In my first internship I prototyped a low-memory-footprint speech recognizer for personal digital assistant devices and a Windows 95 command-and-control application. In my second and third internships I worked on language model adaptation for large-vocabulary dictation.