Siri Ex-Product Lead Who Quietly Sold His Startup’s Assets To Apple Raises $1M For Unsilo, A Semantic Search Engine

Apple’s acquisition of Siri in 2010 gave the company the technology it needed to build a voice-activated personal assistant for its iPhone and iPad devices. A year later, Mads Rydahl — one of the first employees at Siri as its director of product design — sold something else to Apple: a set of patents, nine in all, from a startup he founded before joining Siri.

Today, Rydahl is working on a new startup: a semantic search engine called Unsilo, which is now preparing for a launch in November backed with $1 million from Danish incubator Oei and Scale Capital, a small VC firm co-headquartered in the U.S. (Palo Alto) and Denmark.

Not quite Wolfram Alpha, and not quite like Enigma, Unsilo gives users — initially enterprise customers in areas like scientific research — the ability to find answers to their questions by aggregating data from dozens of disparate sources. The idea behind Unsilo is that the right answers may not be related to the keywords in your search query, but in how those keywords, juxtaposed laterally as they would be in a human’s mind, might help you think of the right answer, essentially applying the concept of natural language processing to search.

“We’re investing quite a bit [of processing power] in NLP of the content we index,” Rydahl tells me. “We have a statistical and semantic approach. We’re not trying to go for Q&A like Wolfram Alpha, nor facets and filters in the way that Enigma does, and although we also index patent data, we’re quite different from,” he says. “They are all, in the end, traditional technologies for indexing and clustering. We have an idea that we can do a completely different type of matching than what’s been done before.” He describes it as extensive query expansion to pick up variations in phrasing and terminology. “Then we detect semantic patterns around the terms that match the user’s query.”

The end result are pages that give users not just lists of documents, but results analogous to the query, grouped by approach, i.e. how they provide answers to the original question posed by the user. You’re allowed to drill down further through the interface, to see refined results.

There is also a cunning, animated visualization that looks like a floating flow chart or organigram, which gives you an idea of how all the different concepts, are literally strung together. The visualization is interactive: if you are searching for common cold vaccinations, and you definitely do not want to see any of the homeopathic-related results, you can remove that concept. Or if that’s what you want to give priority to, you give precedence to results that mention it. The search results reorganize accordingly.

Unsilo search results

Rydahl says that fields like scientific, medical, or legal research, specifically around intellectual property, “are a perfect market to for their technology.” Specifically, he pinpoints the information publishing industry that traditionally makes a business out of providing data research. “That industry is coming apart at the seams, so we are trying to add a new type of value for them,” he says. “They have an opportunity now to make sweeping changes.”

Unsilo is in talks with Elsevier, Thomson Reuters and Wiley, which might become some of Unsilo’s first paying license holders (none have been revealed yet). The medical focus is also helped along by the fact that last year, Unsilo launched, a research tool that helps geneticists discover hidden relationships between genes and diseases, based on evidence extracted from the 20M+ articles in Medline.

Longer term, the idea will be to take the Unsilo framework and apply it to more consumer ends. “Google does a great job on keywords, page rank, source authority, and personalization, but we believe Unsilo can help users move beyond the keyword or the most popular link,” he says. That’s especially apt in areas like research and legal, where you are “not at all interested in personalization.”

The founder, and his Apple story

It’s not often that you come across an entrepreneur who has been involved with not one but two startups that have ended up with Apple. Rydahl is that guy. (And yes, there are others.)

Originally from Denmark, where he is now based again, Rydahl’s resume is a walk through some of the more interesting developments of the Internet, from the early days of publishing making its first transitions to the computer screen, through to core technologies in search and information design, and snagging in consumer games and apps along the way. (In a way, you could argue that what Rydahl is exploring with Unsilo is a combination of all of these: helping publishers migrate their businesses online, gaming elements with the animated category widget, and of course the search technology itself.)

It was the middle part of that experience spectrum, in search and information design, where Rydahl found his life intersecting with Cupertino.

Sublinks, founded by Rydahl in 2000 and closed down completely by June 2012 (after Apple transferred the IP), was one of those under-the-radar companies focused on developing technologies that were potentially more interesting to other technology companies than to consumers or other kinds of enterprise businesses. Considering how closely the media monitors Apple news, it’s surprising that acquisition hasn’t been reported until now. It’s hard to find much on Sublinks today, but one person who had been on its board as far back as 2000 described it like this:

“a nascent search technology and innovation company which holds patents for a number of novel methods within the fields of recommendation systems, knowledge management, and information search and retrieval. The company manages a family of patents in The US and The European Union to protect the rights of its inventions.”

Rydahl described its technology to me as akin to what Amazon does in its product searches. “Recommenders are what Amazon uses when it says, ‘if you like this we’ll suggest that to you.'”

In the early days, he says, this relied on overlaps on items of what two people both bought, which later turned into classes of items that were used to build profiles similar to degree-of-interest trees. Presumably, this is an area that his patents covered as well, and is something that Apple could use, for example, in its App Store searches (among other products).

While Rydahl said that this acquisition was completely unrelated to Siri (it happened about a year later), he also added that, under the terms of that deal, he would not and could not talk about how and if Apple implements that technology today. (And before your imagination runs wild, remember that Apple, like others, holds a number of patents that are not related to any products at all but can be used defensively and for other purposes.)

The Sublinks technology, while perhaps never transformed into a fully implemented application of its own, gave back to Rydahl twice, in fact. Apart from eventually selling the patents to Apple, Rydahl says that it was his work on Sublinks that made him approach would-be Siri CTO Tom Gruber (who is only Siri co-founder who we understand is still at Apple, with Adam Cheyer and Dag Kittlaus both moving on).

Rydahl was employee number-three at Siri, where he reported to Gruber as Senior Director of the team that designed and developed “all user-facing aspects of the product, from concept and early prototypes to a polished, robust and well tested product” — the product that Apple acquired just months after it emerged from stealth.

Although Unsilo has several patents pending on its technology — and Rydahl clearly knows something about how to leverage IP — he says there is a higher purpose: helping people solve problems whose answers may lie in what it already known today, just waiting to be discovered in the right combinations.

“We want to make the world a better place, by building the ‘Google of innovation and discovery,'” he says. “No one else is doing search this way. We have a unique mix of approaches that leverage the best of natural language processing, traditional data mining approaches, and long memory-based linear indexes, and we believe we can leverage this to accomplish lofty goals.”