Clarifai is building the user friendly, independent platform for AI. Recognized by Forrester as “a pioneer in deep learning-based computer vision.” My teams build world class image classification, object detection, object tracking, text recognition in the wild, face recognition, and natural language processing, and do so in a way that others can quickly and cheaply train models with their own data. I’m usually hiring, direct inquires only please.

Blippar is a leading visual recognition and augmented reality platform, selected as CNBC's top 50 disruptors in 2015, 2016 and 2017. Visual recognition provided by an ensemble of search engines utilizing local descriptors, deep features, and deep classification networks.Responsible for the two primary visual search engines. Authored the geometric verification portion of the online system and most of the AWS EMR (big data) index building pipeline.Created and optimized image quality filter used on servers and then phone which reduced server costs for visual search and all deep learning engines by more than 50%, while improving the user experience and extending phone battery life.

Implemented and integrated C++ and python code for stitching and postprocessing of multiple mobile image captures. Architected pipeline changes leading to substantial increase in object recognition. Coordinated updates with domestic UX/mobile team and server team in India. Computer Vision and machine learning: keypoint detection, feature extraction, nearest neighbor feature matching, homography estimation, ransac/huff voting, bundle adjustment, custom clustering, deep learning, scene understanding, observation/validation/hypothesis framework.LiveStroke Developed multiple technologies to make interactive whiteboards more useful. Content-stroke matching for automatic erasure, re-drawing, and summarization of meetings. Developed stroke formats and protocols to support collaboration between tablets and whiteboards and projectors.

Released three mobile applications gathering 2,000,000+ downloads and 20,000+ image captures per day Led the release of the initial ScanPages iPhone application and the White Board Share Android application to provide early document handling on the mobile platform. Architected and built scalable server side image processing that ran mostly unattended for years.

EWRITER Business Application Software Definition and Prototyping for what became Ricoh’s eWriter Solution Business Division. Designed and implemented the “Notebook” application on Ricoh’s prototype eInk based tablet computer. Implemented the first page and stroke model, PDF conversion, document hashes, and template and submission interfaces for the research hardware prototype. Developed course and taught Ricoh developers how to use prototype hardware. Coinventor of algorithms for fast eInk display updates and color constancy, implemented, tested, patented. AUTHENTIDOC™ offered as part of EWRITER workflow services Designed and implemented a secure rolling cryptographic hash model to allow proof of non-tampering with document and stroke content after capture on prototype hardware. Organized the patent portfolio (internally called NEWR), coinventor on many patents in this area. JPEG 2000 Transport protocol standardized and used in medical imagery and satellite imagery applications Invented, and implemented a demonstration of a means to remotely access portions of a JPEG 2000 image. Proposed method to JPEG committee. This was standardized almost unchanged as one of two methods in JPEG 2000 Part 9 (JPIP). JPEG 2000 (now used in space and digital cinema) Ricoh developed CREW (Compression with Reversible Embedded Wavelets) which led to the call for proposals for what became JPEG 2000. Served as a subgroup leader, inventor, and driver for many components in the JPEG 2000 family of standards.