Columbia University in the City of New York

University of California, Berkeley

Building machines and systems for tomorrow

New York, New York, United States

The tools to explore and interact with documents have failed to keep up with the increasing need to understand large (and growing) corpuses of text. Current approaches rely on “keyword search,” ignoring much of what is important about the documents surfaced, and can easily produce overwhelming numbers of irrelevant results.

Merriam combines techniques used in archival research - a focus on meaning and context - with statistical language processing to model the relationship between texts. The new approach allows easy exploration while filtering, flagging, and finding similar documents based on both semantic and contextual relevance. Merriam brings intuitive and insightful interaction to extensive collections of electronic text.

Merriam Tech

Founder/CEO

KFit Solutions is a data science consulting firm that specializes in high-level analytical work for e-commerce, e-publisher, ad, social media, government, medical and financial institutions. Past and present clients include, NYC and CT hedge funds, NewsCorp, The Associated Press, Omnicom, Columbia University, + various NYC based startups.

KFit Solutions

Co-founder

At UnitedLex we are building a lab in the spirit of Xerox PARC with the broad mandate of creating "the  information office and systems of the future." The outlook is for the long-term. As opposed to the quarterly profits or incremental feature wrangling, the thinking will encompass the way in which human interaction with technology and information can be shaped in the next 5-10+ years. Our goal is to shape the future, to invent it or at the least to resurrect some of the great ideas of computing lost in the antic hay of commercialization.

You can learn more about what we are doing here: <a href="https://github.com/UnitedLexCorp/SimpleTalk/blob/master/Mandate_Public_Repository.pdf">https://github.com/UnitedLexCorp/SimpleTalk/blob/master/Mandate_Public_Repository.pdf</a>

UnitedLex

VP Computing

x.ai is a personal assistant that lets you use email to schedule meetings.

x.ai

Data Science Advisor

Co-teaching, with Matt Connelly, "Hacking the Archive," a course devoted to computational techniques in current archival research and development (for more information see <a href="http://www.lse.ac.uk/resources/calendar2014-2015/courseGuides/HY/2014_HY447.htm">http://www.lse.ac.uk/resources/calendar2014-2015/courseGuides/HY/2014_HY447.htm</a>).

The London School of Economics and Political Science (LSE)

Guest Lecturer

Responsible for developing various statistical/engineering solutions in the eDiscovery space.

Paul Hastings

Director of Data Science in eDiscovery

Text processing, NLP, development for the Declassification Engine <a href="http://declassification-engine.org">http://declassification-engine.org</a>.

Adjunct Associate Research Scholar

Advisory role in product development, machine learning, engineering and investment strategies.

Johnson Research Labs

Advisor

Co-teaching the Applied Data Science  <a href="http://columbia-applied-data-science.github.com/">http://columbia-applied-data-science.github.com/</a> course at the Statistics department.

Columbia University

Adjunct Assistant Professsor

At JRL I take lead role in defining, developing and building data science products and solutions that are of interest to the company. The scope of the work is broad and covers statistical modeling, recommendation systems, NLP, data mining, financial modeling, data processing, etc.

Most recently my focus has been on designing and implementing a highly scalable text processing engine for the eDiscovery space. Together with Ian Langmore, I had designed and built a pipeline which processes large volumes of documents of varied type (emails, memos, faxes, financial statements, etc) and returns a relevance classification score. In addition, the pipeline allows for document similarity identification/grouping, social network communication clustering analysis and visualization. The bulk of the pipeline is written in python (with extensive utilization of available multiprocessing libraries) and is wrapped up in a shell script. The classification module is based on an ensemble of random forest and L1-regularized logistic regression, from the scikit-learn library, feature selection is regular expression based and the social network analysis is done via MCL clustering utilizing a combination of metadata and classification results. The pipeline is module and CLI friendly.

Senior Researcher/Data Scientist

Manage the data science team, develop data related products, predictive modeling, recommendation systems, data visualization, model integration, as well as anything relating to data science.

Sailthru

Chief Data Scientist

Cross Commerce Media

Data Mining & Analytics Scientist

Research in the intersection of low-dimensional topology and homological algebra. Teaching of mathematics courses.

UCLA

Assistant Professor

Research in low-dimensional topology, representation theory and homological algebra.

Daniel Krasner