Mick Thomure Email & Phone Number

Portland, Oregon, United States

Portland, Oregon Area

Designed and implemented a model-building pipeline, combining GPU-based modeltraining with a distributed data pipeline. Built a petascale feature extraction pipeline, which decreased cost by an order of magnitude while improving consistency between batch/streaming with a unified extraction code path. Built a system to perform range queries in high-dimensional feature spaces, which decreased cost by $100,000 annually while achieving near-linear scalability.

Portland, Oregon Area

Designed and implemented an optimization framework for the company's real-time ad serving platform, leading to immediate improvements in profit margin and client ad campaign performance. Drove the process from problem formalization, through solution design and simulated validation, and into implementation in a production environment.

Portland, Oregon Area

Helped design and implement a real-time credit decisioning system. Contributed to the company's core machine learning intellectual property. Facilitated sales discussions by conducting analysis of customer data.

Portland, Oregon Area

Prototyped a system for adaptive management of hardware platform resources. Built a mathematical model to predict resource consumption based on user behavior. Demonstrated the model's efficacy on a large set of historical data.

Greater Seattle Area

Helped design a common framework for managing ads-related data. Designed and implemented a reusable provider to serve data in this framework to (1) increase productivity of ads teams throughout the company by decreasing burden of serving data, and (2) increasing the value of existing data by encouraging its reuse. Also prototyped a flexible library for the validation and transformation of serialized data, (3) increasing productivity and minimizing errors in a core company technology. Code written in C++ and Python under Linux.

Greater Pittsburgh Area

Given existing forecasting model used by real customers, performed theoretical analysis of systematic model error and identified impact of possible architectural changes. Ran experiments to validate this analysis, and led discussion of results. Proposed and implemented solutions for most significant sources of error, allowing architectural changes that (1) decreased costs; (2) decreased system latency, allowing system to scale to many more users while remaining highly responsive; and (3) increased system performance, providing much more reliable forecasting estimates to users. Code written in C++, Python, and Bash under Linux.

Improved accessibility of math in PDF documents by integrating MathPlayer, a MathML rendering library, with Adobe Acrobat using Microsoft’s COM architecture. Wrote application to demonstrate usage of new accessibility features for other developers. All code written in C++.

January 2003: Integrated Intuit Quickbooks with web application framework developed in April of 2002, including a network server implemented to handle requests for Quickbooks data. Wrote software to synchronize records between Quickbooks and a SQL data source.April 2002: Developed web application framework in Perl that included a SQL web interface to output formatted views of database records, create web forms by parsing SQL commands, and update tables from posted HTML forms.May 2001: Developed CGI scripts in Perl (under Linux), which integrated with a MySQL database, to perform functions such as storefront order processing and content-management.

Apr. 2000: Ported DAQ software written in January of 1999 to Microsoft Windows NT 4. The code was ported to Microsoft Visual Basic 6, and rewritten to include all devices on the GPIB bus (DAC modules and oscilloscopes) in an object-oriented design.Jan. 1999: Developed software in C (under Linux) to collect data from analog-to-digital converters (ADC) and oscilloscopes using a National Instruments GPIB bus with an X-Windows GUI written in Tcl/Tk.

Phd, Computer Science

Bs, Computer Science