David Jurgens

I developed and invented new video fingerprinting technology to catalog MSN Video clips. I also designed a workflow that added closed captioning to MSN Video clips, which had significant value additions of searching inside the video's spoken text and provided additional content tags for search. Last, I created natural language pattern extraction technology to expand the set of video tags based on website content

While at Sun Labs, I contributed significantly to two research projects:

• Project Darkstar (PDS),
  Mentors: Seth Proctor and Dr. Jim Waldo

  June 2008 - September 2008    I developed a high-performance, transparent caching framework for read-only persistent data. I also implemented an experimental, in-memory transactional database for high-throughput performance. Last, I contributed additional concurrent, scalable data structures for PDS: a deque, queue and linked hash map.

  June 2007 - September 2007   I designed and implemented a high-concurrency, scalable hash map for use in the Project Darkstar transactional application server. The map was designed to support mulitple parallel write operations in a low-latency environment. I developed a customizable logging layer to allow optional transactional semantics on top of existing Java logging. Lastly, I constructed a new profiling framework and an internal testing platform for Project Darkstar with an associated scripting language for rapid test scenario prototyping.

  July 2006 - September 2006   I investigated scheduling algorithms for the distributed, latency-optimized Pro ject Darkstar application server. I co-authored an internal white paper comparing current scheduling algorithms in a latency-bound environment. As a result of the white paper analysis, I invented one new latency-optimized scheduling algorithm and co-invented a scheduling data-structure, both currently submitted for patenting with a goal for later publication.

• Project Vidscape,
  Mentor: Dr. William A. Woods

  June 2006 - July 2007   I developed a distributed infrastructure for analyzing and annotating video segments for later text-based search. During the project, I implemented a new color clustering heuristic for image segmentation and created new representations for shape estimation as a time series. Our work focused on creating an intermediate logical representation for joining video interpretation with the Sun Labs text-based search engine. This project was ended prematurely due to a restructuring at Sun Labs.

I developed software for a five-person start-up based on presenting internet content as a continuously browsable media stream. The focus was to create a semantic clustering algorithm that identified relations in pop-culture data. Relations were combined with a novel, graphical front end for an internet-as-appliance browser. My work resulted in a working prototype that was essential in securing a second round of angel funding.

I was part of a software team that supported both the amazon.com catalog content-build distributed systems as well as software tools used by the business clientele for interacting with the website contents. I integrated additional vendors and data sources into the amazon.com website content, which generated an estimated 750K USD in company savings per year. In addition, I created new software tools for improving developer and business user productivity, resulting in a net savings of 8 hours of developer time and 15 hours of nontechnical work per week. Lastly, I designed a new alarming systems for team-maintained services to increase system stability and reduce downtime.

I investigated fully-automatic extraction of road maps from aerial video by using active contours, or "snakes," to represent both rural and urban roads. We developed a space-efficient spatio-temporal representations of motion estimates using optic flow and tensor fields. This allowed us to invent new heuristics for identifying traffic motion patterns in noisy motion data, which were used to build the final road map of the scene. All research algorithms were implemented using the Intel C++ OpenCV library.

I designed and implemented a distributed framework for testing the feasibility of Real-Time Java in a production environment. Each distributed component was designed to operate in both the standard JVM and a Real-Time JVM. This allowed for dynamic scalability testing as new Real-Time components replaced default JVM components. We used AspectJ to weave in reference counting code extensions to replace existing garbage collection as a test for memory requirements and overhead. The work was presented at the DARPA PCES conference in 2004.

Designed numerous Java software exercises that exemplified course objectives such as concurrency and design patterns for use as course homework. Several projects were used multiple semesters. Proctored one or more lab sessions that included instructing students on course topics and evaluating performance of assigned projects. Created rubrics for grading projects as well as graded submitted assignments and course exams.