A brief list of recent [web] projects
- - The goal of this hack was to make a modern dynamic reinvention of Top Trumps for music artists. Each player gets a deck of artists based on their last.fm artist delivered to their mobile device (currently an iPhone front end, though the server api is such that a front end could be generated later for other devices or an html5 enabled mobile browser). During each players turn, the player selects what they believe is their strongest attribute and a winner is selected based on that stat for both artists value of that attribute. While the winner is being selected an automatically generated mashup of clips from each artist is played resolving toward the winning artist's song. Then the next cards in both decks are loaded and play continues, with the winner from the last round selecting the attribute to compare. Conceived, designed and developed with James Coglan, Jono Cole and Joris Zaalberg at Midem Hack Day Cannes 2011.
- - a service that intercepts the subtitle stream for BBC iPlayer programs and transforms it in various ways, including translation via Google translate and a custom made obscenity engine that adds inappropriate adjectives at grammatically appropriate places (sentence structure analysis uses the excellent NLTK python module). Displays the altered subtitles via a greasemonkey script inline over the program (screenshot from demo, since it's of the obscenity engine, slightly NSFW). Conceived, designed and developed with Daniel Jones and Liam McNamara at Culture Hack Day 2011, where it was awarded "Best Comedic Hack with Potential."
- - a fully autonomous live internet-radio streaming service that uses audience requests to build playlists based on both social and acoustic data. The nominations presented for the requests are also automatically generated with the aim to created playlists balancing novelty, interest and length. This both uses and is a part of my PhD work.
- - The inspiration for this project is a robot solving a maze. Or cleaning a house. The general idea is to construct coherent playlists that traverse social connections from a start to an end song on the Soundcloud with minimal prior knowledge. Playlists are constructed using the social connections of soundcloud and measures of track-to-track similarity. A number of similarities measures are used, including timbrel similarity with timbre analysis from the echonest and a tf/idf vector space model formed from a tracks comments and tags. Conceived and initially built during the 2010 London music hackday.
- - a mash-up providing artist and track recommendations based on what you're currently listening to. Input sources include last.fm scrobbles and up to date BBC radio playlist logs. Output recommendations come from The Echonest and last.fm. Created at the 2010 Stockholm Music Hackday with Mike Jewell.
- - aka pythoniphication of SoundCloud API - some extensions to the existing soundcloud python api wrapper to enable better public page crawling and tools to build a complex network representation of the SoundCloud. Created at the 2009 London Music Hackday. Won the prize from Gigulate for best data visualization.
- - an ongoing effort to bring order to the semantic chaos that is Myspace. Developed and maintained with Kurt Jacobson. This is where a bunch of data from my PhD comes from.