We started off very simply. With an idea to help people who weren't completely comfortable with using a computer. Where did this thought come from? Well, for that, I'm going to have to show you a screenshot:
Now, of course, you could kinda get over your mom typing a URI into the Chrome Omnibox. But no real engineer can bear what's going on in the next picture:
Do your teeth hurt like mine do when you see someone do this?
You see, this is not a computer literacy problem. It's an interface problem and it's faced to varying degrees by different people. It's better described as "an inability to intuitively use an interface". All of us face it when we go to the DMV site, or try to find the best surgeon at a large public hospital. Or try to fill out a visa application form. It's tough!
So what we built was simple. You use your voice to speak to your computer. And the machine takes care of getting the spellings right and figuring out what you're trying to ask and just.gets.you.there to the right webpage. Nothing more fancy.
To do this, we built a series of systems that connected with each other in a modular manner to provide us with great performance, but also to help us change workflows on the fly. We threw in (i.e. spent hundreds of man hours on building) a system that bumps up voice recognition accuracy using the current context (much like your travel agent can figure out most of what you say to him even over a poor internet connection) and shows you a web page that (hopefully) answers exactly what you've asked.
Turns out, the information extraction engine we built as part of the process was very powerful and could be used to extract information for flights, food ordering, shopping, movie bookings and just about anything; and then extract relationships between all the various atomic and elemental units of language in that sentence.
What made our information extraction engine really really special was it's ability to look at what the user had said before - and then derive new insights from this 'context. Amongst humans, this is called 'conversation'. That was truly our AHA moment. Today, nearly 2 years later, we're pleased to be so close to launching our platform. And we're so so excited to be able to release our first API to the public - LeovaTravel for all kinds of travel bookings.
 |
| This is brutal |
Now, of course, you could kinda get over your mom typing a URI into the Chrome Omnibox. But no real engineer can bear what's going on in the next picture:
 |
| Even the monkey can't stand all this monkey-business |
Do your teeth hurt like mine do when you see someone do this?
You see, this is not a computer literacy problem. It's an interface problem and it's faced to varying degrees by different people. It's better described as "an inability to intuitively use an interface". All of us face it when we go to the DMV site, or try to find the best surgeon at a large public hospital. Or try to fill out a visa application form. It's tough!
So what we built was simple. You use your voice to speak to your computer. And the machine takes care of getting the spellings right and figuring out what you're trying to ask and just.gets.you.there to the right webpage. Nothing more fancy.
To do this, we built a series of systems that connected with each other in a modular manner to provide us with great performance, but also to help us change workflows on the fly. We threw in (i.e. spent hundreds of man hours on building) a system that bumps up voice recognition accuracy using the current context (much like your travel agent can figure out most of what you say to him even over a poor internet connection) and shows you a web page that (hopefully) answers exactly what you've asked.
Turns out, the information extraction engine we built as part of the process was very powerful and could be used to extract information for flights, food ordering, shopping, movie bookings and just about anything; and then extract relationships between all the various atomic and elemental units of language in that sentence.
What made our information extraction engine really really special was it's ability to look at what the user had said before - and then derive new insights from this 'context. Amongst humans, this is called 'conversation'. That was truly our AHA moment. Today, nearly 2 years later, we're pleased to be so close to launching our platform. And we're so so excited to be able to release our first API to the public - LeovaTravel for all kinds of travel bookings.
No comments:
Post a Comment