Category Archives: Ideas

Say your name – idea for podcasters

I listen to a lot of podcasts on many different topics. Most of them have nothing in common – except for one thing: they butcher the names of people (cough JavaPosse cough) Those mispronounced names could be of people sending comments, of known people in the community or just some other strangers.

Seems like it would be a valuable service to have a website that people could record their names on. Given that most the podcast speakers know the people to be mentioned one way or another, it would not be to difficult to send them an email request to hit a website that allows to record that name. Then, every time they need to pronounce the name, they could consult the website and save the time to all the listeners with bad attempts and apologies.

And if the website allowed for the name to be embedded in a website as a widget, this could even be a part of people’s calling card to embed their own name on their website. There might even be synergies with identity services such as LinkedIn by exposing person’s name’s pronunciation as a widget on their profile.

The problem of course is how to monetize such a service. Even with short snippets required to deliver the audio for the names, it would add up if the site became popular.

Still, obviously some people think it is worthwhile doing (as a hobby). A search to see if such service already exists turned up a couple of sites:

None did exactly what is actually needed for the podcasts, but something is better than nothing, right? And they did well enough to be mentioned in newspapers or even on TV. Must be something in this idea after all.

Where are all legal computational linguistics resources?

I am frustrated. I know my corpus (resolutions of the United Nations General Assembly) shares a lot in common with biomedical and legal domain. And I can find interesting articles in biomedical domain dealing with similar issues of complex tokenization, long named entity mentions (though mine are much longer), etc. But I see nothing in legal domain.

I have just gone through all of Jurix‘ proceedings as well as all of Artificial Intelligence and Law and all I got is between 2 and 4 articles worth following-up.

There must be somebody actually trying to parse real legal texts and figuring out to deal with complex organisation, people and group names. But all I can see is articles dealing with levels from ontology and up.

There might even be money in it!

One of the crazy business ideas I had was to parse all the web-based terms of use and privacy notices and annotate/crowd-vote them for how bad they are. So, before creating a web-based account, I could check it against database/parser and it would highlight and rate for me passages that I really should pay attention to (e.g. we sell your contact details to every spammer we know ). Since the language of those notices is often ritualistically formulaic, extracting interesting and useful summary would actually be simpler than it looks.

And the business model would center on providing automatic notification option if a notice from subscribed website sneakily changed and became much worse. That way one would pay money for peace of mind that there were no unexpected service rule changes.

Viewfinder Friends – idea for Facebook application

Use case

Photos are inherently-social event markers. We take pictures to remember an occasion and – often – people who were present with us at that point. While most of the photographs are not looked at more than once or twice, the more popular ones become very important in our history.

The same does not happen to other people on our photograph, despite our best efforts to share. Emailing photos is cumbersome and subscribing to Flickr streams requires all people to look at all photos just in case they are in it.

Once upon a time, got a lot of attention for promising to fix that. You would upload your photographs to Riya, train it to recognise your friends and family and it would then automatically find those people in newer uploaded photographs and notify them. Riya even had a basic Social Network, so the photos could be tagged collaboratively.

Eventually, Riya has failed and changed focus to become just another Visual Search Engine. I suspect at least part of the downfall was the single-purpose destination of Riya. You had to register, upload photos, train application, invite other people and do many other basic things with fairly small return on such investment of time. It was much easier to just dump photos on Flickr and let others make an effort of subscribing to your feeds.

Enter the Facebook. It is extremely popular, has all the machinery for registration, adding friends and photos and provides free development API. As it is a platform, rather than a single-purpose destination, it is more sticky than, say, LinkedIn. Succeeding as an application inside Facebook can bring more than a million people to your application.

There are many photo related applications inside Facebook now; more than 400 at the latest count. Most of them however are fairly basic. Combining computationally interesting idea with distribution platform of Facebook could be a wining combination.

Basic business flow

  1. You receive a message from Viewfinder Friends Facebook application that somebody in your network has a public photo with you on it. You can install the application to see which particular photo was tagged for you. That’s the viral distribution method, that most of Facebook applications rely on.
  2. Interested, you add the application. At that point, it shows you the photos from your friend’s profile that he/she has marked with your name. It also presents your photographs and asks you to name people in them. The lookup is all inline and the facebook names are automatically auto-completed, so it does not take much time. The application now spreads to the other friends you nominated.
  3. After a while you come back to your profile and there is another picture with you in it from a friend using the application to tag you. That means he/she have thought of you. Delighted, you keep the Viewfinder Friends in hopes for future signs of attention. The aperiodic, but obviously personal nature of the gift makes the application sticky.
  4. The application does not just show the pictures, it recognises where the people on the photograph are and asks to actually map people’s names to their faces. The mapping can be done by you or by your friends that you identified as being in the picture, but are not yet matched to the face. The distributed marking effort makes the application more interesting and easy to use.
  5. Mapping faces to names also allows to train the application on facial recognition (Riya’s original promise) and later to automatically guess the names of people that the photograph show.  Automatic people recognition will increase return on time investment and will make application more sticky. It will also make the application more viral, as knowing that all but one of the faces are identified and have the application installed, increases the pressure on the remaining people to join the Facebook and install the application.


  1. You take a picture of a friend at a party, waving a can of beer at you. You upload it to your Facebook account and mark it with your friend’s name (say Kevin). Viewfinder Friends pipes up and mentions that the beer in the picture of Kevin is actually delicious Busch Light and there is currently a taste promotion on it run by the distributor in your area. A link is provided to request a free sample.
  2. This information can appear either as application notification, or as a different-color box on the picture. It could show up to you only, to Kevin or to anybody who can see the application. Either way, this is an advertisement that is strongly targeted and is in context. The cost per click (or per transaction) of such advertisements is much higher than for average banner ads.
  3. You go skiing on a weekend and there is a notice at the hotel saying that if you take a picture of the hotel’s front and upload it to Facebook with Viewfinder Friends application installed, you will receive a personal discount of 15% next time you come and stay at the hotel. Happily, you take a picture of yourself, still red-cheeked from skiing, goofing around in front of the hotel with your best buddies. The application recognises the hotel name (or Data Matrix element) from the picture and messages you and everybody else you identified in the picture the discount code. The advertisement costs nothing to the company beyond the initial setup, is created  by happy visitors and is shown to other people in context of the trip. The impact of such an advertisement would be much higher than a stock message.


  • Joyent can provide Facebook hosting that scales up or down depending on demand, making it cheaper to build out the infrastructure. Currently, they even offer some free Facebook application hosting and bandwidth to the limited number of developers.
  • Initial fund for the business can come from one of many Facebook specific funds. The facial recognition part does not have to be implemented until after the application has proven itself popular.