Developing next-generation AI/ML technology that is more robust, interpretable, secure and efficient than Deep Learning.

Developed new machine learning models for text, images and biological sequence data. Focused on improving interpretability, computational efficiency, security against adversarial attack and robustness to input filtering, corruption and perturbation.

Came up with the original idea for the company and raised $34.5M in funding from Redpoint Ventures, Highland Capital and Google Ventures.Hired and managed team of 16 engineers across a wide variety of expertise including computer vision, camera hardware, app development and video post-production tools.Invented algorithms and/or initiated projects behind most of the core IP including custom camera arrays, 3D sound, and computational photography.Drove IP protection strategy including the filing of 7 provisional patents and 4 utility patents.

Managed team responsible for creating the only new web search engine with substantial traffic since the initial launch of Google. Grew headcount from 7 to 25 with world-class talent.Led transformation of the product from a small-scale demo to a state-of-the-art web search engine with full stack of crawling, indexing and machine-learned ranking.Invented algorithm to automatically classify queries with high accuracy against over 1,500 distinct categories, as showcased in Blekko's tablet search app Izik
.

Developed proprietary GPU-based computer vision technology and helped evaluate potential investments.

Managed over 30 direct reports across 10 projects in the core areas of Online Ads, Web Search, and Systems Infrastructure, including the following company-critical technology: ◆ AdWords Real-time Auctions: Massive-scale statistical models that decide which web search ads to display by computing real-time estimates of their expected revenue ◆ AdSense Contextual Targeting: Semantic analysis algorithms that determine the topic of web page content in order to accurately target contextually relevant ads ◆ Web Search Index Selection: Data-driven optimization to select which web pages should be searched in order to provide users the best possible results ◆ Web Search Spelling Corrector: Probabilistic algorithms behind the popular “Did you mean…” context-sensitive spelling suggestion feature ◆ Map/Reduce: Massively-parallel distributed computing paradigm that powers the vast majority of Google’s data processing (see en.wikipedia.org/wiki/MapReduce)Project contributions were both technical (e.g., architecture, algorithm design and in-depth data analysis) and managerial (e.g., staffing, planning, mentoring, reviews and promotions). Invented novel patent-pending algorithms for web search indexing and semantic modeling.

Developed a multi-platform toolkit for programming the massively parallel video-processing component of the Silicon Optix Realta SoC that powered many high-end video productsDesigned SIMD C, a C-like language with custom syntax for supporting unique SIMD (Single Instruction Multiple Data) features of Realta video processor.Wrote optimizing compiler for SIMD C with 40K LOC and over 150 C++ classes and created interactive scripting environment for testing, profiling and debugging SIMD C programs based on Java, the Eclipse SWT GUI widgets, and embedded Jython.

Joined Plumtree during aquisition of my startup Ripfire.Led effort to integrate Ripfire's search technology into Plumtree's flagship corporate portal software used by over 500 customers worldwide including Ford, Pfizer and Citibank. After Plumtree’s subsequent acquisition by BEA, Ripfire search engine was sold as BEA Aqualogic Interaction Search.

Founded and ran the daily operations of Ripfire, an enterprise software startup company.Developed innovative, high-performance search engine software that successfully competed against offerings from well-established competitors, such as displacing market leader Verity as the site-wide search technology at Wal-Mart. Designed and coded initial version of Ripfire’s patented search engine, raised venture capital financing and served on Board of Directors, and hired engineering, marketing and sales team.Coordinated sale of company from initial presentations to final negotiation.

Developed critical algorithms and software components for Arithmos’ flagship ASIC video display driver, which was designed into flat panel monitors from manufacturers such as Hewlett-Packard. Arithmos was acquired by STMicroelectronics in 1999.Invented patented image processing IP, including dithering and scaling algorithms, wrote software simulation of entire ASIC design to verify Verilog code and created GUI developer tools and designed micro-controller software architecture.