startup interview Luke Stanley

Looking around for an interesting startup can be difficult. Before they're really under way, there's not much sign of them and unless you talk to the principals you won't learn much because of NDA's and a general fear of saying the wrong thing.

In this case, John Littler was staying at a friend's place in North London for a night and there were Luke and Alex from getting ready for a presentation the next morning. Conversations ensued! Here is a chat with Luke Stanley, their CEO. What's it about? In some respects, easing the web experience: In other respects, ushering in a new era - the stuff that intelligent agents were supposed to do but, for various reasons, mostly haven't as far as the ordinary user is concerned. is a startup looking for angel investment and coders who want to collaborate... Python coders who don't mind getting dirty in the Windows API and using C++ when needed.

Did your project result from a sudden epiphany or more develop through a series of ideas?

A bit of both. I had an epiphany after a series of ideas developing, in response to various cravings I had, to solve various annoyances with how we use computers. The epiphany was that, the problems were solvable, and that they can be solved by giving computers a place to think and reason, that is more than just files and folders storing documents or other data. It's about metadata.

As I understand it, in basic terms, your app has a look at the user's desktop (+ other files?) to form some kind of context and then parse whatever input - chat, messaging, web - for keywords. These can then be used in a variety of ways such as generating things that might be interesting including ads. Am I on the right bus here?! Please tell us more. As far as the user's desktop and files are concerned, what has been done to ensure the user's privacy?

Sort of. So as well as related Google results, YouTube videos and images, being available without having to retype what you're looking for, we're interested in related products and services being shown when the user wants them too.
An example in the mockup we're working to, is that my friend Mark stole my happy tree friends tshirt. Bastard. But, if the person I'm talking to decided they wanted to buy one, maybe for themselves or both of us, a non-intrusive picture with a link to do just that should be in the MiniBrowser window somewhere, in the same way a related Happy Tree Friends YouTube video should be easy to find.

So, like regular desktop search, in-fact: complimenting regular desktop search, we pull together data that gives us an idea what the user wants, or may want in the future. In order to predict what the user wants, we have to share some of their experiences, and model the interests based on that. So, we've developed a way of getting what web pages the user is loading up in Firefox, getting text from Instant Messenger conversations, and we can reuse existing Desktop Search data too.

So we have this data about what the user does, but what do we do to ensure the users privacy? Well, we store the information locally. So you hold on to your information, and not some server somewhere on the Internet, that you don't know about. We see privacy issues as software bugs, that can be solved with a little careful thought.

E.g: what if you wanted to backup your ThoughtTrail data online, so when your computer breaks or whatever, your computer isn't lobotomised? Well, a solution would be to automatically encrypt it when putting it online, with cryptographically secure methods. So you could have it download it and get your new computer would decrypt it using your encryption key, and then remember all this information about what stuff you like, what stuff annoys you etc. so ThoughtTrail can let your computer can help you out again.

... continues below ...

The attractions for commercial concerns seem quite obvious but what are the selling points to get users to download or use a web app? Actually, which is it?!

That's a good question, it's actually both a web app, and software that the users download. We let users blog from their conversations, so that part is a web app. But the core of ThoughtTrail, the ThoughtTrail Platform is something the users download as part of MiniBrowser or Instopix.

The selling point, ok: ThoughtTrail was built to solve frustrations I had, the main one being: Why am I running my web browser manually and googling this text, when the text is right there on my screen already, why can't I just get related information when I say I want it? Rather than messing around with my web browser? So we get related information up without the hassle of messing around in the web browser. We make it just a few clicks, and certainly don't want the user to have to retype anything. If you had a personal assistant or friend looking at your screen, they would be able to TELL you related stuff easily, but computers can't do that yet: well, not without stuff like ThoughtTrail. We provide a place for the computer to think about, and reflect back upon what you do, so that it can save you the hassle by doing stuff for you.

I think I heard some mention of AI at some stage. How is this used?

Well, there is AI at several stages. What parts of the conversation define it's meaning? It's an open-ended question. We try solving it with a general purpose phrase finding API, but we're working on adding various clustering techniques to find out how all this information interelates.

This is a really interesting bit I think. Extended a little bit, it could give a whole website some contextual meaning or allocate a personal relevance score of some kind. How close are we to this? Right now, from what I've seen, we haven't got much further than word frequency tag-clouds ... which can be some sort of guide to meaning, but a fairly limited one. Semantic parsing looks promising - what do you think? Is this level of "understanding" beyond your needs?

It is certainly not beyond our needs. The most accurate model of what the user wants if made by the greatest array of data you can get on them, The better it is, the more useful you can be.

An extreme way of looking at it is: some say the universe we live in could be us living inside the mind of a great Artificial Intelligence, just considering different possibilities. The more realistic the simulation, the better your ability to make more intelligent predictions to help the user out.

It's well known that aggression is effected by temperature. When temperature goes up, crime rates go up until they are too high and people become more lazy. If the temperature was one degree higher of lower during the Cold War, one of those knife edge moments could have been a final moment. Modeling temperature and psychology could, for example be a reasonable part of trying to predict market reactions and so on. Ask a weather derivatives trader. ;)

So that level of understanding is certainly not beyond our needs. We do in-fact have the ability to find personal relevance scores, we can find out how interrelated any two given topics are. With the ThoughtTrail Platform It's simple to iterate upon that and find out how interrelated a whole list of topics are. Specifically, how strongly it interrelates to the users own interests. There are so many different, interesting 'thoughts' of different possibilities to look into, that it becomes apparent, that a platform is truly needed to make the best use of our computers.

You're tying things together with Python I think, or is it more than that? Why Python?

We're using Python for the plugins because it's beautiful. I haven't seen code more beautiful and maintainable than Python code. We could have used Javascript, and it would make the download smaller, but it would make programming for ThoughtTrail less fun, and make adding to it more of a chore. A lot of other AI researches love Python too. We use Python for the core right now, but we're probably going to replace chunks of that with more lower-level code, to make it even faster. We still want Python for writing plugins. All you have to do is open up a text editor and hack away. You don't need to download a compiler. When people grow up with computers, trying to pull together all these contextual clue's: we want them to be able to use Python to refine the computers intelligence, and share their plugins back with the community so everyone can use it.

The process sounds quite CPU intensive but I guess it can't be. Do you use indexes and such to keep it manageable?

Yes, we do use index's to save on using lots of CPU and memory. We use SQLite to store the Thoughts objects, which is pretty efficent. Firefox recently switched to using SQLite too. Pre-processing data allows us to find related information a lot faster when we come to reuse it. For example, Google doesn't have to look at every word on every webpage it ever looked at, in order to find out what webpages to show you, it just has an index of words and the words you search for are pulled from that index, and it already knows what webpages it has that contain those words. In-fact, a lot of searches Google has, have been done before, so they don't even need to do that, a lot of the time. We like how Google has optimisation at every level. In some ways ThoughtTrail is inspired by Google's architecture.

You're looking for venture capital funding at the moment. Is the UK a good place for this? Is the current financial climate proving a factor? US news from that sector seems to be more optimistic than news generally!

Actually, we're more interested in angel investors at this point. It's cheaper for me to meet angel investors in London, (from my home in Coventry) than it is to get a flight to San Francisco or New York, staying there, and meeting them there. That's just how the money situation is :) I would prefer to be in San Francisco again right now, and it would certainly make finding angel investment easier.

What are the next steps after you receive some funding?

We want to outsource a bunch of stuff and take on some permanent hires, roll out the platform for developers to build on, move to the Bay Area, release a series of products to make computers smarter, etc.

Thanks Luke.

No problem :)

Back to top


post to Delicious Digg Reddit Facebook StumbleUpon

Recent on Mstation: music: Vivian Girls, America's Cup, music: Too Young to Fall..., music: Pains of Being Pure At Heart, Berlin Lakes, music: Atarah Valentine, Travel - Copenhagen, House in the Desert

front page / music / software / games / hardware /wetware / guides / books / art / search / travel /rss / podcasts / contact us