With additional writing by Josh Catone.
This morning saw the first five startups at the TechCrunch40 event hit the floor and presented to journalists, investors, and early adopters. The first session focused on search startups. It’s interesting to note that one third of the presenters at the TC40 are from outside USA. Below are quick reviews of the first five companies to present.
Powerset
Powerset is a natural language processing-based search engine that raised $12.5 million a year ago to create what many have touted as a “Google killer.” Company founder Barney Pell said this morning that interacting with today’s search engines is like talking to a 2 year old. What Powerset does to change that is index the web semantically and extract meaningful relations.
They are still very early in the process of perfecting their technology, but today they will be opening their Powerlabs beta site that lets users search within pre-defined areas (such as a quotes database from Wikipedia). Powerset will solicit feedback from users to help refine their technology. People will be able to contribute ontologies in which they are specialized, which will allow them to win “search karma.” More search karma means earlier access to Powerset’s bleeding edge technologies.
At present, Powerlabs has 3 specialized verticals based on restricted subsets of Wikipedia: Quotes, PowerMouse, Business. Business and Quotes are self explanatory, while PowerMouse searches Wikipedia entries for connections between topics. For example, you could search for “Yankees” and “home run” and find home runs related to the Yankees.
Cognitive Code
Cognitive Code makes an artificial intelligence platform called SILVIA (or, Symbolically Isolated, Linguistically Variable, Intelligence Algorithms). Their platform provides a technology for starting human-like conversations with machines and an intriguing interface to start dialogue. For example, you could tell SILVIA to open the most recent file on your computer and it will process your command via voice recognition, understand the meaning, and execute your request. Their linguistic variability technology allows them to understand what you mean — you can give your command in many ways, but their software will still understand your intentions.
Cognitive Code says that they have already signed licensing deals with toy companies, but there are many more opportunities awaiting them.
CastTV
Founded by a husband and wife team, CastTV is a video search play that is attempting to index every video on the web. CastTV indexed videos from many sources, including user generated video sharing sites like YouTube, mainstream media like CNN, and paid download services like iTunes. They deliver their results using an impressive clustering technology that lets you drill down results by date, relevance, source, price, and more. When you search for Britney Spears, it asks you whether you are looking for her music or sex tapes, for example.
FAROO
FAROO is a P2P search engine that has no central crawler or index (so something like Wikia’s Grub?). Their demo was blazingly fast and brought to mind my recent open source distributed Google clone article.
FAROO says their P2P search technology can save traditional search companies $1 billion. On the other end of the spectrum, the advantage for users is “revenue sharing and attention ranking,” they say. They are committed to sharing 50% of all search revenues with users and the quality of their ranking is based on the time you spend on these sites; in other words, they follow you. This brings privacy questions into play.
For them, distribution is going to be a problem — without people, there is no value in their technology. Wikia is known to be very open to small acquisitions, however, and already dabbles in this field, so could be a potential suitor.
Viewdle
Viewdle is another video search engine. Viewdle uses a facial recognition algorithm to search for people within videos. The main problem is that people need to be in their database to be covered, and so far they only index celebrities. Viewdle already have a deal in place with Reuters, who Om Malik thinks will end up buying them. Marrissa Mayer asked Viewdle how they plan to scale when their database enlarges to many people, and their answer was “contextual analysis,” which will allow them to recognize faces after the environment.
Full disclosure: Emre Sokullu joined search startup Hakia as a Search Evangelist in March 2007.