I like watching sports but normally can't keep up with them all, so like any good bandwagoner, I come out in full force during the playoffs of each sport. Something about watching teams and players when it's their season on the line brings the best entertainment. The two I keep up with are Formula 1 and hockey. My favorite team is the Lightning, and I guess Max Verstappen, but I'd rather see a good race than Verstappen winning by a monster margin.
In March of 2020, my girlfriend at the time (now fiancee) embarked on a two-week Irish adventure only to land in Ireland with COVID starting to take full effect. We tried sticking it out and spent the day in Dublin going to the Guinness factory and then on a whiskey tour. We had stewed kidneys that I still dream about, and then the next day, we took a train across the country to Galway with the intention of renting a car and traveling city by city back to Dublin over the next two weeks. Then the talks of not letting people back into the US started, and we thought it was best to get back, so we spent our two days in Galway and then took the train back to Dublin to get the first flight home. We have every intention of returning and eating those kidneys again, and I'll finally get to drive on the left side of the road legally.
My role at Vianai is working on our H+AI Platform, specifically on backend services. I integrate third-party libraries or create Vianai libraries that we then expose for client use via the platform (HTTP API) or various camel routes. While I don't deal directly with customers, a lot of the customer feedback gets refined into additions they want to add into the system. Many customers come with the challenge of managing large amounts of data in a performant way. This in itself creates many interesting challenges with leveraging multithreading, complex SQL queries, and looking at third-party libraries that may have already fixed these problems and implementing into our platform. No reason to reinvent the wheel...
As of late, the focus of my work has been on inferences and risk. An important part of our product is allowing MLOPs engineers to understand what is happening with their deployed models, so we need various monitoring tools to look into what could be going right or wrong. We do this in two ways: looking at the output of the predictions or labels and monitoring various accuracy metrics. We also look at the features or inputs that the user gave to do the predictions and compare those to past inferences. This allows us to see if the underlying data is changing in such a way that a model's training may need to be re-done so that it can take into account this drift in data. We have dashboards that display these to the users so that they can make informed decisions and maintain their models to be as accurate as possible.
The most exciting part is the challenge of thinking about the product as a whole and not just one specific problem. We're building a platform that can support multiple industries, model frameworks, databases, messaging channels, etc. We want the ability to plug seamlessly into whatever configuration and industry-standard applications they are currently using and allow them to run from day one of installing the platform.
I think customers appreciate our attention to detail and our turnaround time on feedback. Iteration is key, and the best way of doing that is getting working proof of concepts or software in front of a client so that it can be adjusted accordingly. The ability to touch and mess around in the software allows them to think of things they may not have thought of before by creating requirements or writing down their thoughts. That iterative process is what leads to the best solution for everyone involved.