A mechanical engineer, computer scientist, and professor emeritus at the Federal University of Minas Gerais (UFMG), Nivio Ziviani still possesses the vitality of his younger years in his manner of speech, even as he nears 74. As one of Brazil’s pioneers in undergraduate training in computer science, he attained accomplishments in his lifetime that few academics in his generation can claim, not only training professionals and producing research, but also applying that research outside the campus walls.
After completing doctoral research in the 1980s at the University of Waterloo, in Canada, he was stricken by the “entrepreneurial virus”, as he likes to say, creating five startups throughout his career. His successful businesses soon attracted notice from the market—one of them, Akwan Information Technologies, was acquired by Google in 2005, leading the multinational to locate its research & development center for Latin America in Belo Horizonte.
“Since then, Google has invested hundreds of millions of dollars in Brazil. This was made possible by the way we were able to harness research to create technology, and apply that technology in the real world to build businesses and create high-quality jobs,” he says. “Brazil’s universities need to be an engine of wealth creation through innovative business.”
Ziviani met with Pesquisa FAPESP for an interview just before the novel coronavirus pandemic broke out. In the following months, the interview was supplemented by telephone interviews and exchanges of messages. Below are selected excerpts from the interview.
Field of expertise
Algorithms, information retrieval, and artificial intelligence
Institution
Federal University of Minas Gerais (UFMG)
Educational background
Bachelor’s degree in mechanical engineering, UFMG (1971); master’s degree in computer science, Pontifical Catholic University, Rio de Janeiro (1976); PhD in computer science, University of Waterloo, Canada (1982)
Published works
44 articles in scientific journals; 14 books (8 as coauthor); 121 papers published in conference proceedings
You had a dramatic experience during childhood. How did this affect you?
As a child, I had polio, and this had a big impact on me. I underwent multiple surgeries, the most aggressive of which was in 1960, when I was 14. In an operation to correct a shortened tendon, the doctors had to cut the bones in my foot. I spent 90 days in a cast. This had a profound, and probably a decisive, effect on how my life unfolded from there. Because of this limitation, I wasn’t competitive in collective sports. But I always enjoyed riding my bike. At 16, I fitted it with an engine. I liked tinkering with machines, gears and linkages. This led me to study mechanical engineering later on.
But although a trained mechanical engineer, you spent your entire career as a computer scientist. Why so?
I have always enjoyed motorsports, and that’s where my interest in mechanical engineering came from. As an undergraduate student, I managed Antônio Lúcio [“Toninho”] da Matta’s kart team, which won the Brazilian championship in 1966. But during my second year of college, the School of Engineering at UFMG bought a computer, an IBM 1130, one of the first in Brazil. I soon began working as a programmer, and this eventually drew me into the computer science community.
What was it like working with computing at a time when there were only a handful of computers in Brazil?
There were only a few hundred of them. They were huge machines, each filling an entire room, and their computing power was much less than that of today’s smart phones. They were difficult to program. I completed my training in mechanical engineering in 1971, all the while working with programming. The next year, a friend of mine, Ivan Moura Campos, a mathematics professor at UFMG, invited me to take a course in computer programming. And that’s how I became a part-time lecturer.
Were you then already considering an academic career?
No, it hadn’t crossed my mind. After earning my degree, besides my role as a part-time lecturer at UFMG, I worked as a programmer at the now-defunct National Bank. In 1973, the government launched a project to develop technologist-level undergraduate programs at five institutions: UFMG, PUC-Rio [Pontifical Catholic University of Rio de Janeiro], the Federal University of Rio Grande do Sul [UFRGS], the Campina Grande campus of the Federal University of Paraíba [UFPB], and Centro Paula Souza, in São Paulo. In March that year, I was invited to work full-time in helping to implement the program at UFMG. It was an incredible experience.
In what way?
We started on a blank slate. We formulated the entire curriculum, and had to hire professors. It was a unique, two-year, intensive training program divided into quarterly periods—at that time, all university programs were divided into semesters. The same curriculum was implemented at five different schools.
Part of the money that Victor Ribeiro and I earned from the sale of our first startup was donated to UFMG
This was one of Brazil’s pioneering undergraduate programs in computer science.
Yes. The first admission exam was held in mid-1973. I served as program coordinator at UFMG for two years. Because the programs were using a novel approach, there could be pushback from the associate dean for undergraduate programs, so the Ministry of Education [MEC] thought it best to have program coordinators manage the funds directly, which meant I reported directly to the dean.
What is your assessment of computer science programs today in Brazil?
They’ve come a long way. Brazil leads Latin America and has world-class teaching and research. Our graduate programs are competitive, we train highly competent professionals, but unfortunately we suffer from brain drain. Our top talents, many of whom have gone to college with taxpayer money, often end up working abroad. Much of the credit for the quality of Brazil’s computer science education goes to Carlos José Pereira de Lucena, a professor at PUC-Rio who I regard as the father of computing in Brazil. He created the first master’s program here, and played a crucial role at CAPES [the Brazilian Federal Agency for Support and Evaluation of Graduate Education]. He also influenced me to go to Canada for doctoral studies at the University of Waterloo, the country’s most prolific startup incubator—BlackBerry smart phones were developed there.
When did you earn your PhD?
I defended my doctoral thesis in 1982. When I returned, professor Lucena, then head of the computer science department at CAPES, invited me to serve on committees evaluating graduate programs. In 1984 I returned for a three-year postdoctoral program at Waterloo. While there, I was appointed as Lucena’s successor. I served for two terms as head of the department.
What are your chief interests in computing?
My early interests were algorithms and information retrieval [a branch of computer science that deals with the storage of documents and retrieval of the data they contain], but I later branched out into artificial intelligence [AI]. This change of course will be better understood in the context of the industrial revolutions that have happened in the past. The first, in the nineteenth century, resulted from the invention of the steam engine; the second, in the early twentieth century, was ushered in by the internal combustion engine and electricity. These are general-purpose technologies that are a part of our everyday lives. Artificial intelligence is also a general-purpose technology, and should have the same dramatic effect that the combustion engine and electricity had on society. The third industrial revolution occurred in the 1990s when Tim Berners-Lee created the World Wide Web. The internet can be thought of as a repository of texts. And people needed mechanisms to retrieve those texts, which led to the development the first search engines. In the 1980s I began working with natural language processing, an interest I acquired during my doctoral search.
A number of search engines emerged in the 1990s and 2000s.
Yes. The most iconic, of course, was Google, developed in 1998 at Stanford University. In 1993, my supervisor, Gaston Gonnet, developed one of the earliest search engines for the internet, which he later spun off into one of the largest IT [Information Technology] companies in Canada, OpenText Corporation. He created a total of 10 startups. Waterloo is Canada’s hot spot for startups, and while there I was infected with the entrepreneurial virus.
Akwan developed strong R&D capabilities and cutting-edge technology through collaboration with a world-class research group at the university
So when you returned to Brazil, you were already infected…
In a way, yes. In 1985, Gonnet came to Brazil to deliver a course on algorithm design. We met up and he suggested we create a company specializing in text-search algorithms, the basic mechanism of search engines. I kept the idea in the back of my mind. Ten years later, while lecturing at UFMG, one of my students, Victor Ribeiro, developed a search bot that could crawl the internet and retrieve pages from target web servers.
Was it a search engine?
No, just a software program, or bot, that had this function. Creating a search engine was not an easy task back then. Nobody knew how to do it. After he completed the program, Ribeiro came to work with me at the Information Processing Laboratory [Latin], which I had created in the 1980s. One day, he watched as his colleague browsed online bookstores on the internet. He would visit each store, and use the store’s proprietary software to find a book. Victor thought: “Why don’t I use my bot to search all online bookstores at the same time, and return the aggregate search results to users?”
In other words, he envisaged a search engine specifically for books.
Yes, to find books from bookstores anywhere in Brazil and around the world. This is known as a metasearch engine: using third-party engines to retrieve pages from each business’s server, aggregate the search results, and display them to users. This was the starting point for an entire family of metasearch engines: BookMiner, for books; CDMiner, for CDs; SoftMiner, for software, etc. Metasearch is currently a thriving industry. AirBnb, Trivago, and MaxMilhas are all examples of metasearch engines. We created our engine at a time when nobody knew how to do it.
How did this engine become your first startup?
It all started at Latin. The Miner family of Web agents was a success, and the department’s network soon became overwhelmed. The number of engine users was doubling every 30 days. Because we were no longer able to host the system on the university’s servers, we made a deal to host the Miner family on UOL. At the end of 1998, we won an iBest award [Brazil’s most prestigious internet ward] as the most popular and most high-tech website. The Miner family was then being maintained with money out of my own and Ribeiro’s pockets. He decided to quit his job to dedicate himself to the business, a must for any entrepreneur to be successful. I continued my work at UFMG. In early 1999, I opened a newspaper and was surprised to find an article by business columnist Luis Nassif, titled “The Miner family.” He praised our engine as a “cutting-edge technology born out of a university project.” I immediately called two of my colleagues—Ivan Moura Campos and Guilherme Emrick, an investor and the founder of Biobrás, an insulin manufacturer—and convinced them to invest in the business. And that’s how Miner started.
How much did they invest?
The amount was not publicly disclosed but it was a relatively small investment. Ultimately, UOL became very interested in the business and acquired it in June 1999. I am not at liberty to disclose the acquisition price. Ribeiro would go on to work as Chief Technology Officer at UOL for several years. This was one of the first success cases involving a spinoff from a university project.
In 2011 UFMG implemented a framework that facilitates technology transfer to startups
Is it true that part of the proceeds from the sale of Miner were donated to UFMG?
Yes. When we created the search engine at Latin I wanted it to be co-owned by the university. But there was a lot of bureaucracy involved and this proved impracticable. Ribeiro and I agreed that if the business were successful we would make a donation to UFMG. So we donated R$100,000, or the present-day equivalent of around US$90,000.
What was the money used for?
We suggested dividing the money into three parts: one part for modernizing the laboratory, another part for creating a graduate library for ICEx [Institute of Exact Sciences], and the third part for creating a permanent scholarship program for scientific initiation in computing. This money was placed into an endowment fund managed by FUNDEP [UFMG’s Foundation for Research Development] and the interest went to the Miner Latin scholarship program.
Soon after selling Miner, you created another startup that was later sold to Google. Could you tell us more about this venture?
In September 1999 I met Berthier Ribeiro-Neto, a professor here at UFMG, at the University of California, Berkeley. While there, we made plans to create a general search engine instead of a metasearch engine. Two months later, we launched Todobr. We founded Akwan Information Technologies as the search engine provider, with companies like Odebrecht, Editora Abril, and the iG and UOL portals as clients. The tool soon expanded to other countries, like Chile and Spain.
Could you describe Akwan’s business model?
Unlike Google, which was a media company, Akwan was a technology provider. Todobr was a huge success. Akwan had a strong R&D [research and development] department and offered cutting edge technology. This was enabled by collaboration with a world-class research group, like the one today at UFMG’s Computer Science Department.
So Akwan was a spin-off from UFMG?
It was created in the room that now houses the Artificial Intelligence Laboratory. We soon moved it off the campus to a building near UFMG, but continued to collaborate closely with the university. I was determined to keep everything aboveboard, so I devised a special corporate mechanism.
Could you elaborate on this mechanism?
We created a corporation, Akwan S.A. [Sociedade Anônima], and donated 5% of the shares to a foundation linked to UFMG, FUNDEP. We described the university as a partner in the business, although it wasn’t in the strict sense of the term. We often attended international conferences also frequented by people from Google, where we discussed our experience at Akwan. This was what put us on their radar. Google had just completed its IPO [Initial Public Offering] in 2004, and was looking to establish a presence in Latin America. Near the end of the year, we received a visit from Google’s vice president of engineering, Wayne Rosing.
He came here to Belo Horizonte?
Yes. But because I was in Europe, he was met by Berthier, Akwan’s executive director, and by founding partners Guilherme Emrich, then a co-owner of investment consultancy FIR Capital, and professor Ivan Campos, the two of whom had previously invested in Miner. Akwan had two other partners, Marcus Regueira, from FIR Capital, and Alberto Henrique Laender, a professor at UFMG.
How did the meeting go with Rosing?
We met at a hotel in downtown Belo Horizonte. We were reluctant to take him to Akwan because we were afraid our engineers might recognize him. But he was adamant about visiting the company, while we did everything we could to stall him. At one point, he said: “Just a minute, let me contact Mountain View [where Google is headquartered, in California].” He asked us to download a Nondisclosure Agreement [NDA]—a standard procedure before initiating negotiations—and signed it. My partners then agreed to take him to visit the company. To prevent Google’s interest in Akwan from leaking, we told the researchers that he was just someone visiting our offices. But one of our engineers had attended the International World Wide Web Conference in New York a few months prior, and recognized him: “This guy is from Google!” The very thing we were trying to prevent from happening, happened—the news leaked.
Personal archive
Ziviani (right) in a meeting at UFMG with Google founders Sergey Brin and Larry Page (both left)Personal archiveWhy was it important to keep Google’s interest confidential?
The negotiations were very incipient and there were still lots of loose ends. Rosing was taken back when he saw first-hand what Akwan’s operations were like and the quality of our software. Under an NDA, we started a negotiation process that lasted more than six months—the deal was closed on July 19, 2005. Berthier and Regueira, the ones negotiating on our side, went to California several times and met with Google cofounder Larry Page, who was directly involved with the deal.
Six months is a long time.
The negotiations took so long because of all the details involved. Hiring engineers was one of them. Because labor costs in Brazil can be prohibitive, at Akwan we had decided to hire through labor cooperatives. This was a common practice at the time, but involved a labor risk. Another issue was that Google wanted to acquire 100% of the business, including FUNDEP’s 5% interest. They required a document from the foundation’s Board of Trustees and the State Office for Foundations attesting that the sale of foundation assets to an undisclosed buyer and for an undisclosed price would be lawful.
What was the selling price?
I’m not at liberty to comment. But it was big money. The contract had 10 pages of provisions requiring that neither of the parties disclose the price. It was Google’s first acquisition outside the US.
And Akwan would later become Google’s R&D center for Brazil.
Not only for Brazil, but for Latin America. The center was set up in Belo Horizonte, with Akwan engineers, and then opened sales offices in São Paulo. Google has invested hundreds of millions of dollars in Brazil. This was made possible by something simple, but which people don’t always give enough importance to, which is harnessing research to create technology, and applying that technology in the real world to build businesses and create high-quality jobs. Brazil’s universities can be an engine of wealth creation through innovative business. The country’s science output jumped from 0.8% of world research output in 1996 to 2.6% in 2018. But it hasn’t generated GDP [Gross Domestic Product] at the rate it should.
In other words, Brazil’s universities generate knowledge, but not wealth.
Not as they should. Public universities have a moral obligation to use their scientific knowledge to create wealth through innovative ventures, giving back to society the money it invested in training high-quality human resources. In addition to training top professionals, they need to produce wealth as they do in developed countries.
And why isn’t it the same here?
The answer is complex. There are a number of legal obstacles. Everything I was able to achieve before founding Akwan was a struggle, even though the management at UFMG was very supportive. In 2011 the university implemented a legal framework that facilitates technology transfer to startups. In return, UFMG receives 5% of the company’s registered common shares. UFMG has the same rights as other shareholders, but not the right to vote. This detaches the university from the company. With these types of shares, if the company generates liabilities or fails, the shares vaporize, but with no other consequences for the university.
You started other businesses after selling Akwan…
Yes, I set up another startup four years later with a former doctoral student, Edileno Silva de Moura, now a professor at UFAM [Federal University of Amazonas]. He was the creator of the Todobr search engine and had worked as a software engineer at Akwan. The startup, called Zunnit, specialized in recommender systems, a subclass of information retrieval systems. These systems essentially recommend news and information related to a user’s interests. But Zunnit didn’t perform as expected, because back then, in 2009, the newspaper industry was already flagging and the market was limited. We decided to shut down the business and open a new one.
Which company was that?
The company was called Neemu and specialized in search engines for e-commerce. I cofounded the company alongside Moura, Altigran Soares da Silva—another professor at UFAM—and four students from Manaus. With strong information retrieval capabilities, the company provided search technology to e-commerce giants like Americanas and Shoptime. In 2014, 30% of Brazilian e-commerce companies used Neemu technology. The following year, Linx, one of Brazil’s leading players in retail technology, decided to enter the e-commerce business and made an offer for Neemu. We completed the sale for R$55 million, in September 2015. In March the following year we created Kunumi.
What does it do?
It solves complex problems using artificial intelligence. For the past four years we have used AI to assist companies in applications like predicting supply and demand, identifying anomalies, and analyzing lending portfolios. Our clients include large corporates such as Itaú, petrochemical company Braskem, Coca-Cola, Porto Seguro, and pharmacy chain Raia Drogasil. Kunumi also does R&D for the benefit of society. We recently did research to help doctors and policy makers better manage the COVID-19 pandemic. Our research was reported in an article I coauthored with my colleague, Adriano Veloso, and published in the medRxiv repository in June.
What is your routine like today between your business and academic work?
In addition to my work at Kunumi, I am a professor emeritus at UFMG. I mentor students and participate in programs at centers of excellence. Last year I was surprised to receive an invitation to join Petrobras’s Board of Directors. CEO Roberto Castello Branco wanted a scientist with an entrepreneurial bent and expertise in artificial intelligence. One of my missions on the board is to assist the company in its digital transformation. I was happy to accept the challenge.
