Inside Architext

The Herring takes a look at Architext, a Silicon Valley start-up founded by six recent Stanford graduates, and their journey from the garage onto the Internet.

The Herring Reporter

March 1995

The sports car Mr. Kraus was referring to belongs to Vinod Khosla, the general partner at VC powerhouse Kleiner Perkins Caufield & Byers (KPCB), who just sank almost $250,000 of his firm's money into the recent startup. "The guys at Architext have solved some very difficult math problems," Mr. Khosla told The Herring. These math problems were part of the breakthrough software development work being completed by a band of five burrito-eating computer science majors from Stanford University. The end-product appears to be a next generation search-and-retrieval text engine which is uniquely suited for the burgeoning online service market. "They have really made major breakthroughs in a number of areas that are important in the online world. Their software functions seamlessly in a highly distributed environment, allows concept-based queries, and performs automatic hypertext linking. It's really quite amazing," says Mr. Khosla.

Quite amazing indeed. The "official" story of Architext Software's founding, is that a group of friends, five computer hackers and one political science major, who first met each other in their freshman dorm at Stanford University, were sitting around one evening at Rosita's Taqueria in Redwood City, CA, discussing their after-graduation options. "We knew that we didn't want to work for any big company--we wanted to work together and do something entrepreneurial," said Mr. Kraus, as he sat talking to The Herring in the middle of the carpeted living room also known as the company's conference room. "I'm not sure whether or not it was at Rosita's that we actually developed the specific idea of what we wanted to do, but that's what we've been telling people," Mr. Kraus explains, with the true entrepreneur's twinkle in his eyes.

When, exactly, the idea for their new company actually popped into their heads is perhaps irrelevant by now, but what is important is the assumption this young team made about the emerging Internet market, and their opportunities to play a significant role in its development. "We knew the Internet was growing and it was becoming a useful tool for people to communicate and identify important information," explains Mr. Kraus, "but when we reviewed the existing crop of software tools available to Internet users, we couldn't find anything capable of managing the vast amount of text available. If you look at services like LEXIS-NEXIS, they are very primitive." What they did find was a set of different text-search tools that were incomplete, and failed to incorporate the ability to use concepts to help people locate the specific type of information they might be looking for. So they loaded their bookbags and went off to do research at the Stanford library on what methodologies might be applied to solve the search-and-retrieval and browsing function not yet available on the market, and they started developing tools that focused on this area.

During the research phase, their whole project almost got derailed when the founders took a road trip and went to DB/Expo, where they found several companies trying to solve the same problems. "We were very naive and believed all the marketing hype being put out by companies such as Conquest, that were claiming to have the kind of software we were trying to develop, so we got really bummed," says Mr. Kraus. Eventually, the Architext team realized that these companies did not actually have working products, and thus got their first examples of the oldest product line in Silicon Valley, known as "vaporware." Revivified, the technology team, led by Graham Spencer, the company's chief technologist and primary technology guru, spent the summer after graduation hacking away. "It's all kind of a blur," recalls Mr. Spencer. "All I remember was that Joe started scheduling demos before we were actually finished, so that kind of kept us pushing ahead."

When they first came up with the concept, a flurry of e-mail was exchanged amongst the founders on exactly what approach to take to develop their product. They ultimately had two choices. One was to take the natural language approach, which has been implemented by Oracle in its Context product line, and the other was a statistical approach, which was being heavily advocated by Mr. Spencer. "I was arguing that there were ways to achieve better performance without needing an artificially intelligent computer or relying on Boolean methods," explained Mr. Spencer. The natural language approach is cumbersome, because it requires a big dictionary of terms, and a lot of information about different parts of a sentence and grammatical structures, so that concept-based queries can be initiated. Oracle claims, for example, that it took them one hundred man-years to build the knowledge-base for its Context product. But, as Mr. Spencer explains it, the statistical approach uses neither a dictionary, nor complicated algorithms, to determine what phrases mean, but numerically tries to get the same information by looking at the proximity and co-occurrence of key words and phrases. "We believe that the natural language approach will not be viable for another four or five years," says Mr. Spencer. "The Context product has made some great strides toward a natural language understanding, but I believe it is still very inefficient and very brittle, because it takes an enormous amount of resources and support to make it work," he adds. Not surprisingly, Mr. Spencer's logic prevailed, and the company converged on the statistical approach, writing most of their code from the ground up on an old Sun workstation. "It took easily 15 minutes to compile our code, so our lack of computer power was really our biggest stumbling block in the early days," reminisces Mr. Spencer.

The company got its first big break when they were discovered by some executives from InfoWorld, the weekly news magazine and conference company based in Foster City, CA, and owned by International Data Group (IDG). Amanda Hixson, the IDG executive who has worked most closely with the Architext team, told The Herring that she had spent a lot of time looking for the best online service tools, and determined that "the boys" had the hottest product in the market. "The key differentiating factor of Architext's product is that it isn't merely a search-and-retrieval engine, but a package that also includes automatic hypertext linking, which is a handy feature for browsing, and dynamic subject grouping, which helps condense a large list of documents into subgroups of related topics," Ms. Hixson explained. What exactly Ms. Hixson and IDG are doing with the Architext technology is not yet ready for public consumption, but Mr. Kraus did disclose that InfoWorld provided some highly appreciated early service revenue for the young startup, right around the time they were surviving on rice, beans, and moxie. "We put together a deal with InfoWorld by July of 1994 to build a prototype system for them, for a little less than $100,000, and we needed some cash, so it was a pretty good mix of motives," says Mr. Kraus.

In a somewhat roundabout way, the InfoWorld deal eventually led to Architext's next round of funding, which came in the form of equity capital from the professional venture capital community. IDG's executive vice president, Stephen Coit, also turned out to be a partner with Boston-based Charles River Ventures. But he wasn't willing to fund the deal unless he could "find a West Coast investor," presumably to keep an eye on the kids from Silicon Valley. Without a business plan in hand, Mr. Kraus started randomly calling west coast VC firms, so Architext could close the deal with Charles River. "We essentially brought the VCs over, showed them our technology, and said 'Isn't this neat?' But they didn't go for that," he recounts. What the Architext team would learn from this experience is how little imagination most VCs have when it comes to looking at raw startups. The VCs were all looking for a presentation that would show how their firms could gain a "20X" return, which was a math equation that even this young team of brilliant computer scientists couldn't solve. "We got pretty jaded by the whole experience, and I got very sick of talking to the VCs," says Mr. Spencer. "We even got a rejection letter from Brentwood Associates even though we never gave them a proposal," he added. "They even misspelled my name," chimed in Mr. Kraus, as all six founders erupted in laughter. They keep the Brentwood letter pinned to the wall, to remind them that perhaps the real geeks aren't software hackers after all, but the VC entrepreneur-wannabes driving around Sand Hill Road in their Porsches and other ultimate driving machines.

It wouldn't be until December 1994 that Architext would find VCs enterprising enough to provide them with the seed money and advice they needed to build a big-league company business plan. "We met Geoff Yang from Institutional Venture Partners, who then introduced us to Vinod, and all of a sudden we were on the VC fast track," says Mr. Kraus. Mr. Khosla told The Herring that Architext was a classic KPCB deal, because it represented a good market idea and a good technology. "KPCB really has a long history of working with raw startups," he says. And indeed they do. Genentech, Tandem, and dozens of other startups were actually born in KPCB's offices. Even Mr. Khosla's own startup enterprise, Sun Microsystems, was funded by KPCB back in 1982, when he and his three other cofounders were only 27 years old."The Architext guys remind me a lot of the original Sun team," reminisced Mr. Khosla. What impressed the Architext team immediately about Mr. Khosla was that he bought them a $4,000 hard drive, right after his first meeting with the founders. "It was like pre-sale support. We just met the man and he bought us a hard drive. That went a long way with us," says Mr. Kraus.

Geoff Yang, who matched KPCB's $250,000 seed investment in Architext, has a similar philosophy. "I saw six really smart guys, who were extremely entrepreneurial, and pioneering a technology applicable in a hot new field," says Mr. Yang. He adds that the Architext team meets the profile of "the new breed of entrepreneur we like to invest in--young, half-nerdy, half-artistic, and not necessarily out of the computer industry."

Since December, Messrs. Yang and Khosla have been working very closely with the company, trying to figure out Architext's business model and putting together their business plan. "What sold me was the fact that both Vinod and Geoff essentially offered to go to work for us, and help us figure out how to leverage our technology, and they have definitely lived up to their commitment so far," said Mr. Spencer. "Vinod has the high-level strategic vision, and is full of ideas. Geoff is very thorough, and helps us a lot with details and contracts," adds Mr. Kraus. Presently, Architext and its VC investors have agreed to spend the next six months prototyping different business models. "We don't think it's wise to position Architext as a search-and-retrieval engine company, because they'll end up in bake-offs with a lot of established vendors," says Mr. Yang. As the first part of their strategy, Architext plans to introduce a next-generation Internet navigation service within a month. "By combining our technology, such as hyperlinking, subject-grouping, and automatic abstracting, with personalization, we think you get a uniquely efficient service for locating information on the Net."

Once the long term plan is established, KPCB and IVP have pre-agreed to invest another combined $1 million in the company. And assuming the relationship continues, they will jointly put up an additional $1.5 million in 18 months. The next lesson the young team will learn, perhaps, is what it's like to turn over more and more of the ownership of their company to investors. While the company would not disclose the terms of its VC deal, it is fairly standard practice for VCs to gain controlling interest by the second round of financing. And once VCs are in deeply enough, they're known to exert some of the control they acquire to keep young companies on track, and out of low-budget Mexican diners. The company's founders are excited about having some hard cash, and the business experience now at their beck and call. In a milestone move, Architext prepares to go from a home to an office environment, and the founders can suddenly afford to go see a movie during evening hours, instead of always having to go for the matinee. For now, as Vinod Khosla and his founding partners at Sun Microsystems used to say in the early days, Architext's future seems so bright they have to wear shades.

The Short History of Architext Software

Fall 1989
Joe Kraus, Graham Spencer, Mark Van Haren, Ben Lutch, Martin Reinfried, and Ryan Mc Intyre live in the same freshman dorm at Stanford University. Mark was the dorm's Resident Assistant.

Spring 1990
Kraus, Van Haren, and Mc Intyre form Where's Julio?, a jazz and funk band reminiscent of Oakland's Tower of Power.

February 28,1993
Spencer suggests, over a low budget Mexican dinner at Rosita's Taqueria in Redwood City, that the six join forces and build software that helps people more efficiently search through large bodies of data on the Internet.

June 1993 through February 1994
The Architext team works for three months compiling research on different approaches to new search-and-retrieval and browsing techniques, and completes initial series of software tools. Team members were all holding other income-producing jobs during this period.

May 1994
Company makes presentation of their software to IDG, and wins service contract to complete work on a secret online service project. The $100,000 deal was actually signed in July, just as the company was moving out of a Palo Alto garage, into the Cupertino home of three of the founders.

August 1994
Architext creates their own Web page, complete with virtual garage, and MarketText, a program which allows entrepreneurs to create marketing slogans, with as much panache as a high priced ad agency, at a fraction of the cost. Find at http://www.atext.com

December 1994
Kleiner Perkins Caufield & Byers and Institutional Venture Partners agree to invest $500,000 in seed money. KPCB partner Vinod Khosla, and IVP partner Geoff Yang roll up their sleeves and start helping company write its business plan.

March 1995
Architext founders make cover of The Red Herring.

Copyright 1995 RHC Media, Inc.