Thought Leaders in Artificial Intelligence: Francisco Webber, CEO of Cortical.io (Part 3)

Francisco Webber: In the end, it translates to a situation where we could ask a person to manually annotate something like 50 documents, and the system would pick up the characteristics like the typical content, position, and context where a paragraph containing this would appear. We could demonstrate that with 50-100 examples, the system could be trained to perform even better than a human would, especially if the human has to annotate thousands of documents.

This is a problematic situation. That was one out of the ten use cases that we confronted there. They decided to license the technology at that time even in its crude form. After that, there came a bunch of other large players who allowed us to work on specific use cases in the domain of finance, chemistry, and car manufacturing. That was something that we saw quickly. There was no obvious domain in business terms where what we had would apply. It was rather a kind of use case that became the vertical.

Sramana Mitra: What were the big use cases? You mentioned one with this bank, what other use cases were recurring?

Francisco Webber: In the end, doing this practical pioneer work ended up with the use case of extracting information from contacts or documents in a more general form. A second product is for SaaS tech data which includes messages, tweets, emails, and social media posts. These are stuff that comes at a very high frequency which has typically poor language.

You want to find out if people are complying with site rules or if people are complaining on the customer service line about a certain product. We call this message intelligence today. This is a system that allows you to specify by example. It would be like this, “Here are 50 example emails of how people complain, if you ever see something like that, send it to the complaint desk.”

You can to some degree, also do this statistically. I say to some degree because as far as I know, there are no strong systems out there. The problem is that every filter that you implement can cost you the price equal to a few servers per month to just run that one classifier at the required speed. We found out that the blocking situation in the market was the price per filter that you have.

If you could drop the price per filter to a couple of cents, then suddenly people in large enterprises who do have those hundreds of those filters can start thinking about doing it. The third component that is offered as a product is a search system. It is fundamentally like Elasticsearch, but it is not using statistics to index the data. It uses semantic representation.

You can do things like typing in an English query into a Spanish collection of documents and you will still find precisely what you are looking for. The semantic representation is independent of the actual language, it is only dependent on the meaning of whatever has been expressed in that language. These are the three products that we offer. We pride ourselves in each of those products in many different business areas. 

This segment is part 3 in the series : Thought Leaders in Artificial Intelligence: Francisco Webber, CEO of Cortical.io
1 2 3 4 5 6 7