• Data mining algorithms for Collaborative Filtering (with some NNMF)
• Distributed computing and storage (including map-reduce technology such as Hadoop)
• Some minor work with CUDA early in it's history. However I now much prefer working directly with CUBLAS when I need large/dense matrix calculations. (So much so that I wrote my own Python/Numpy bindings for it.)
• Many NLP-related machine learning techniques including but not limited to information retrieval, stochastic language modeling, text generation, word sense disambiguation, entity recognition, etc.
• Some minor bioinformatics work.

• Wrote an automated multi-language classifier for an Arabic search engine company.
• Wrote a browser-based ontology editor (using Python/Django) which is now a top-level product for the company (Jager).

• Automated build and deployment processes (Ant)
• Version/source control management (CVS, GUI diff utils, etc.)
• Requirement/enhancement/defect tracking system (Bugzilla)
• Custom modifications to our tool-set provide automated code-to-requirement traceability (for example: CVS/Bugzilla Integration)
• Implemented an iterative, incremental development process which minimized both customer and company risk by providing a constant, visible stream of working features
• Implemented numerous automated development reports designed to provide complete USAF transparency into our development history, current progress and future goals
• Moved (carefully, slowly) our customer's interest away from a waterfall development mentality to many of the "agile" development principles

• Common, linked user interface used on all screens
• Database abstraction away from JDBC code in servlets and JSPs (similar to JDO, before it existed)
• Collection of reusable HTML generating software components, allowing rich GUIs to be developed quickly (many AJAX-like using hidden iframes and DOM manipulation, before AJAX was invented as an acronym and long before libraries of such tools existed)
• Common screen and data level ACL security implementation, with associated management tools
• Automatic database schema versioning, verification and upgrades
• All code reorganizations were done live, meaning they coexist(ed) with existing, working designs, as to afford the design change without incurring substantial delays and loss of functionality

Data Knowledge Objects (DKOs) - An ORM Done Right
DKOs are a loose Java port of the Django Query API. I originally pitched this as an internal project at work (to replace our POS Hibernate setup - are there any non-POS Hibernate setups?), but was turned down (not unreasonably: a large project + outside the scope of the department I was in). So (with permission) I implemented it as open source outside work, and once it proved itself brought it back in as an external library. Currently in widespread internal use and the default for new code accessing the database.
http://code.google.com/p/nosco/

DelugeFS - A Small-Scale Shared-Nothing Distributed File System
DelugeFS is a distributed filesystem I implemented as a proof of concept to show that many of the elements of a shared-nothing large-block distributed filesystem are already well established (and would be easy to tie together). Built on Python, FUSE, Bittorrent, ZeroConf and Mercurial.
http://code.google.com/p/delugefs/

DeSQL - A Log-based SQL Tool
For work I needed a Linux-based SQL util that kept searchable logs of all commands run (and optionally results given) and would let me quickly post-filter results. This was the result.
http://code.google.com/p/desql/

Deseb - Django External Schema Evolution Branch
This is an implementation of schema evolution for Django. It is a stand-alone derivative of the original 2006 SoC implementation of schema evolution. Funded by the Google Summer of Code project. (currently unmaintained)
http://code.google.com/p/deseb/

gPapers
An open-sourced, Gnome based digital library manager (for managing research papers).   Briefly mentioned in Nature.   Currently unmaintained.
http://gpapers.org/

PyStatCVS
A Python reimplementation of the semi-abandoned StatCVS utility. (My major motivation for this was StatCVS's lack of branch analysis and extremely long run times.)
http://code.google.com/p/pystatcvs/

NOTICE: These are all in varying states of maintenance and completion at any given time.