The Oracle part five

There’s very little about me that’s remarkable. I had a happy childhood, my adolescence was normal, and I guess you could say that I peaked during my early adulthood, so I have to wonder what is the cause of the situation I find myself in. Maybe you can help me understand, so let me explain what I mean.

My parents were both smart people, and seems like that passed on to me. I never had any trouble with the school curriculum, but that also meant that I never really put any effort into it. Maybe this was the reason why I never felt any accomplishment from the things I was doing, at least not from those related with school. When it was time to go to university, I let myself be influenced by others because I didn’t really know what I wanted to study, which stemmed from having no idea about what I wanted to do. Since I was well above average in maths, the school counselor suggested that I enrolled in a mathematics programme, and so I did. I kind of liked algebra and calculus and geometry, at least at the level they were taught during high school, so it didn’t seem like a bad choice. And from what I had read online, if you were able to complete the programme successfully and graduate from a good university, you actually had good chances of finding a decent job, either by going into the private sector or continuing with research in the university system.

So, mathematics it was. Now, as you can imagine, it was a relatively small world, as not many people had the calling to go and tackle the monster head on, as there are few subjects that instill fear in a human as much as mathematics does. I never understood that fear, as for me it was always a different language for talking about what is around us, it was always something that was grounded in reality even when you were talking about the eigenvalue of a matrix, or the minimax of an hyperbolic paraboloid - there’s a reason why those are called saddle points! This did made me feel somewhat disconnected from my peers at university, because they were mostly interested in very weird abstractions and theorems, the purity of the science, trying to probe deeper and deeper into the intricacies that made up the language, but seemingly straying away from any sort of actual useful application that might come from what they were doing, as that was never a worry. This difference made me understand that I was someone you could call an applied mathematician, a distinction that my thesis supervisor once mentioned during one of our meetings. He was the one that suggested that my thesis should be about the intersection between computer science and mathematics, and I followed on that thread.

An efficient implementation of reverse depth-first search was the pompous title of my bachelor thesis, but in fairness it did what it said on the tin. My implementation was truly better than pretty much all the existing ones, and the computational speedup was appreciable, which is a characteristic that is appreciated by the private sector, especially tech companies that are trying to push the boundaries of what software can do. So it came as no surprise that by the time I was supposed to start searching for a job, I already had twelve offers, some from companies you for sure have heard of. In the end, I chose an internship at this startup that was working on natural language processing, which baffled both my family and my thesis advisor; it’s true that the salary was less than one third of the second-lowest offer, and while the promise of equity needs to be taken into account, money played absolutely no role in my decision. Rather, it was him.

I guess you would understand if you knew him. He is, without a shadow of a doubt, the most intelligent person you will ever meet, and not only that, but he carries himself in such a manner that can only be described as electrifying. He commands without having to ask, because what he says is clearly the most logical and sane solution to the problem at hand, and he is clearly only focused on solving problems, no ego, no vanity, no presumption. I once watched another person correct him, and he just smiled and said “You’re absolutely right!”, and continued without batting an eye, just integrating this new information into his train of thought and chugging along. It was, honestly, inebriating. So, I could only choose to hop on for the ride.

The internship, though, was honestly quite boring. Those 3 months were mostly just spent around shadowing other engineers, trying to get acclimated to the workings of this company, and understanding the width and breadth of what they did. Some of it was quite easy for me, some of it was absolutely impenetrable, as they were dealing with such a large problem that it required extremely specialized knowledge in many disparate areas, some of which didn’t make sense to me until I got the explanation of why they played a part into the larger scheme of things. And the general architecture of what was being built there was indeed quite interesting. I won’t bore you with the details, but from what I could see and evaluate, they were light years ahead of anything else that was being done elsewhere.

Once my internship was about to end, I was called into an impromptu meeting with the leadership: him, the CTO, and someone from HR. Some pleasantries were exchanged, but lasted maybe ninety seconds, probably even less. The CTO cut right to the chase, Look, we like you, and seemingly everyone you shadowed during these three months only had good things to say about you, your academic record was also somewhat impressive, so we’re extending you an offer to join us full time, but be warned that the expectations from that kind of role are completely different than from what you did until now, I’m sure you have noticed that you were mostly dilly-dallying around, but from now on, you are bound to actually deliver, and while we do believe that you’re fully capable of doing it, it must be said right from the start that this is the expectation, what do you think?

I was a bit taken aback by this, for a reason I can’t quite put my finger on. I knew that this company wasn’t looking to exploit me, as I could see that everyone had a reasonable work-life balance, no managers breathing down anyone’s neck, and people seemed generally content with what they were working on. This was in stark contrast with pretty much everything I knew through second- and third-hand evidence about the existing state of the labor market, where the opposite was the norm. But, likewise, I never heard anyone talk about this in such a straight manner, this need to deliver, and could only wonder what it all meant. And I guess something gave away my thought process, because he jumped in after two seconds.

Look, son, I understand that this might all sound quite strange, but you have to understand our position, what we’re doing here isn’t exactly your run-of-the-mill white collar job, not even the normal IT job, and I’m sure you’ve noticed it, so that’s why we need to apply some pressure in this regard, and either you accept that and commit, or you don’t and go for something else, it all depends on you, and either way is fine with us even if we would greatly prefer to have you on board as we’ve noticed that you might be a good fit to a project we have running, and we’re having some troubles with, and your combination of skills and personality seems to us to be quite interesting, have you ever heard of lemmatization, it has been quite difficult for our system to infer the meaning of a word from the context, and we noticed that the algorithm you developed during your bachelor might be useful for cracking this egg.

And I answered with the truth, I have no idea how to deal with this problem, but I’m more than willing to give it a shot. The smile on his face, more akin to a grin than anything else, was the answer that the HR lady needed to present me with a contract and some terms, You can read them at home but we expect an answer from you until the end of the week, since we need all the firepower we can get to get this project back on track. And so I went home and had a read, and from what I could see there was nothing there other than the normal legalese, and my father corroborated this and urged me to take this offer as he had heard through a friend that works in IT that there was a lot of buzz around startups working in this field, and he was clearly proud that I was finally taking an interest into something that might lead me to have a good life, at least financially, because the new offer they gave me was indeed more attractive than what the internship would lead one to believe. And so I wrote an e-mail letting them know that I’d be accepting their offer.

This new position, in a new team, with people I had never met even though I thought I had shadowed every department in the company, turned out to be exactly the opposite of everything I had experienced before. None of the answers to the tasks that were given to me came in easy, so I had to truly apply myself. But with this also came a sense of accomplishment, that I was making an useful contribution and making things move forward towards a common goal. It’s hard to explain unless you experience it yourself, it’s like you have broke through some sort of invisible wall and all of a sudden a new expanse is right in front of your eyes. It was truly exhilarating, this new feeling, and I got hooked. I was doing extra hours in the office and at home, even though no one ever asked me for that, just because I truly cared to deliver. I was doing exactly what they asked me in the interview.

Still, the problem we were facing was indeed hard to crack. We tried many different things, from well known algorithms to more out-of-the-box solutions, but it was honestly so complex that we were making little to no headway. For the software to understand the meaning of the word, it needs to understand the context; but it seems like it couldn’t get it from the context if it couldn’t comprehend most of the words that composed said context. Humans do this quite naturally, but the software couldn’t. And then it hit me: we were looking at the problem all wrong.

Our approach was to build a meaning-map of all words in a given context, and from that have the software create a representation of what a given block of text meant. But this was a very intensive and costly approach, as we were checking each and every word and then getting tripped up on words that could change meaning depending on context - hence the work on lemmatization. This was highly expensive, and even though it worked at first, once the efforts to scale up took place, it all unravelled. So, I suggested to do the opposite: let’s build first a representation of the block of text and only then update said representation by looking only at the words that could change meaning, and update the representation if we found anything worth changing.

This approach took a while to implement, as it implied a complete reworking of not only our parts, but as well as the collaboration of several other teams. Thankfully, once I explained my approach to the leadership, they all seemed to agree that it was at least worth a shot, even if they were unsure that it could work. From the start of our work until the first prototype was concluded, it was 8 months of grind, but by the end we could show that it was working: our benchmarks now had a remarkably high score in all the pain points that were present until then, and the computational costs for running this new approach dropped so much to become almost negligible.

With this new system now in place, everyone got aligned to build on it, and the speed at which we were able to build an MVP that was so far ahead of anything else was remarkable, as in only 6 months we were able to deliver demos to many clients with the new functionalities. Internally, this new MVP had a very funny name. We decided to call it “The Oracle”.