Artificial Intelligence

The PhD was moving along, I was starting to work on topics related to the thesis, and we were able to establish a day-to-day routine in the new city. At the same time, several colleagues from DCC/UFMG (including my brother) were being recruited to work at an American company that had set up an office in Belo Horizonte. It was an artificial intelligence company called Intelligenesis. The company’s goal was to build an artificial general intelligence (AGI) system and apply it to the financial market. One of the ideas was to set up an investment fund managed by AI. The company’s big focus at that moment was building the infrastructure needed to train and run that artificial intelligence. There was also a heavy focus on natural language processing, which basically means reading and interpreting text written in human languages.

Intelligenesis paid in dollars, which meant the salaries were higher than local ones. There was also the promise of stock options, but that depended on the company’s future and the possibility of an IPO. At the time, few people understood that whole stock options thing. In Brazil, it wasn’t common for people to have that as part of their compensation. Intelligenesis was seen as one of the .com companies of the dot-com boom in the 1990s and early 2000s. Today, it would be called a startup. The company was heavily dependent on investment to continue operating, but the focus on managing investment funds was a big draw for Wall Street investors. That’s also why the company’s headquarters was in New York, right near the New York Stock Exchange and the World Trade Center.

One thing became very clear: Gisele had no desire to go back to Lavras. Lavras was a small rural town and, outside the university, the job market for someone with a master’s degree in computing was practically nonexistent. Combining our wish to take a different path with the fact that Intelligenesis was growing and hiring, including in networking, support, and security, I started talking to them about the possibility of working for the company. Since I knew practically everyone at the Belo Horizonte office, the conversation was easy. There was an interview with Cássio, who was the head of development, and with Stephan, who was at the same level and handled, among other things, security, networks, and infrastructure. The conversation was good and they decided to hire me.

Now I just had to leave the University of Lavras. Since I had taken six months of leave, the agreement was that I would have to pay back the salaries I had received during the paid leave. We didn’t have much savings, and the amount to be paid was pretty significant for us. I managed to negotiate payment in installments, which really helped ease the pressure on our budget. Since Gisele was getting a scholarship and the salary I was going to earn at Intelligenesis was higher, we’d be able to pay the installments. With that, I managed to fully cut ties with the University of Lavras and began working remotely for Intelligenesis from Campinas. And so ended my first experience as a public servant in Brazil.

One of the first tasks I had at Intelligenesis was improving the security of their internet connection from the New York office. They had only a proxy between the internal network and the internet, so the idea was to migrate it to a DMZ and set up a Linux firewall using ipchains. I started the work by designing the network and doing the initial ipchains configurations. Meanwhile, other people were hired for the company’s infrastructure area, including a manager, who became my direct boss and who had been one of the system administrators at DCC/UFMG since the time of the “telnet incident.” Another hire was Elias, who already had experience with Linux firewalls and gave my initial configuration a pretty good polish.

Shortly after, Intelligenesis called me to New York to spend two weeks there and not only get to know the company’s infrastructure better, but also help make the necessary improvements, including to internet access. The new boss had decided to swap out the equipment connecting us to the ISP without relying on the ISP’s support, because someone said their support was very slow. The good part of that trip was going alongside Kenji, who also worked at the company at the time and was my adventure partner around the Big Apple. Intelligenesis had an apartment to host employees, which was close to the office, but it was fully occupied during the first week of our trip. So they sent us to a hotel, which seemed fancier and farther away than necessary. The upside was getting to know the city a bit more. In the second week, we were transferred to the company’s apartment.

As for internet access, we planned it out and set a date to stay at the office overnight to make the switch. Right at the start, we detected a problem: when we connected the new equipment to the ISP’s network, even using the same IP address, there was no communication with the internet. There were four of us trying to debug the problem, and nothing worked. Until one colleague, a Russian who had been hired at the New York office, proposed not only copying the IP but also the MAC address of the old equipment. The MAC address is an identifier used by network equipment to communicate via the Ethernet protocol and, normally, it’s fixed for each network interface. Until that day, I didn’t even imagine that it was possible to change the MAC address of a network card, but the Russian said it was possible on Linux, for certain kinds of hardware. It was a bit of a desperate option and a good Brazilian-style hack (gambiarra, as we say in Brazil), even though it was proposed by a Russian. We did it and it worked. At the time, I also didn’t know enough about ARP cache behavior or gratuitous ARP, which could have helped us understand the stale IP-to-MAC mapping on the ISP’s side and avoid that hack. Another option I learned later was the possibility of flooding the MAC table of the switch, making it behave like a simple ethernet hub.

Since everything seemed to be working, the boss sent two of us home to sleep. Since the corporate apartment was nearby, they’d call if they needed help. The next morning, I woke up a bit late and headed to the office. When I got there, the boss had a sour look on his face. He said he had called the apartment several times and no one had picked up. I didn’t understand, because the phone hadn’t rung.

Either way, we had a bigger problem to solve: the new internet connection was working for almost everyone, except for the CTO and founder of the company. He couldn’t connect remotely via VPN. We spent hours trying to understand what was happening and analyzing packet captures, but with no sign of progress. We made changes to routing and firewall configurations, but nothing worked. We were already at the point where the only option would be to revert the whole change when, suddenly, the CTO managed to connect the VPN.

I can’t explain what happened with that network, but I learned to avoid unilateral changes (i.e., changing something without coordinating with the other side), and that computer networks have an almost esoteric component: things happen and no one knows why. As for the apartment phone, we later discovered that another employee felt entitled to unplug it from the wall to keep it from ringing at night, without telling anyone, and we were left hanging.

After that scare, the following days were calmer and uneventful. On the day of my departure, I went to the office in the morning to work until it was time to head to the airport. I remember that at one point the person in charge of organizing trips saw me there and asked when I was going back to Brazil. I replied that it would be that same day, and he was shocked: nobody had told him, and he hadn’t booked a car to take me to the airport. He checked the flight, told me to run back to the apartment for my bags, and said he’d send a car to get me. When I got back to the front door of the building with my bags, there was a limousine parked in the street. I thought it would be in the way of the car that was coming to get me and kept waiting. Until the driver stepped out and called my name. The car service hired by Intelligenesis had no more regular cars available and ended up sending a limo to pick me up. I got in and we headed to the airport, just me in that huge car. The car even had a minibar, where I found a glass bottle. I asked if it was water, and the driver laughed and answered, with a vaguely Russian accent: “It’s vodka, but you can drink it if you want.” I immediately dismissed the idea; it would only make my thirst worse.

Despite the rush, I made it to the airport in time and boarded the flight back to Campinas. Time to get on with life, take care of the family, and continue the PhD.

My timeline

Artificial Intelligence

Links: