About PathAI and what they do
Yeah, PathAI is a AI powered tech startup based in Boston, our whole mission is sort of improving patient outcomes with a powered pathology. So if you're not sure what that means, basically, you know, pathologist gets a stain or slide of some blood or tissue and then they normally look at it on a microscope and make some clinical diagnosis, what we do is we try and train machine learning models to be able to classify every single cell or piece of tissue on an entire stain. And then with that, we can provide all sorts of tooling for pathologists to sort of understand better kind of what's going on in that particular stain. So rather than have to look around, we can generate heat maps and overlays for them. And we can generate reports about sort of what the model seats to give them a little bit more insight.
Details on PathAI's engineering team
So PathAI is engineering team consists of about 110 ish, I think at the time of recording this engineers ranging from machine learning engineers to DevOps to SRE to sort of full stack, we sort we split the company a little bit into two halves, one's the product engineering side, which is about 65%. And they focus on sort of developing applications for end users, like the slides here that we make, or the clinical trial services platform that clinical trial might interface with, or that contributor network that a pathologist wants to come in and annotate slides that we've digitized. And then the other half the ML half, which is the side that I'm on is composed are comprised of machine learning engineers, who sort of focus on building and training and evaluating models. And then ml ops engineers, which is where I work, which is a lot like DevOps, but sort of with a little bit of ml sprinkled on top right, we care a little bit more about sort of the ecosystem that Emily's like to work in and research and deploy things. So if you're not familiar with that, it's effectively, you know, systems around the training, lifecycle, evaluation generating of assets. How do you deal with model governance and drift detection, stuff like that? So that's ml ops base, specifically?
Cool projects engineers get to work on
Yeah, there are a lot of really interesting projects that we had to work on. One, for starters, that interests me personally a lot is, you know, we we ended up outgrowing AWS, we had to build our own data center. So we have our own high performance computing data center filled with 700, of the most cutting edge GPUs and this huge parallel file system and, and lots of really interesting network challenges. So there's lots of you know, you get to work with physical hardware a little bit if you're interested, which I think is a fascinating space is where my background is in. But then there's all sorts of other things, you know, there's problems around, we have this massive amount of data, petabytes of slide data that we have to move around and feed to these GPUs during training and inference. How do you sort of deal with optimizing the movement of all of that data? That's a pretty common struggle that you deal with when you're working with these very large systems, and you want them to be low latency and high performance, how do you get this data move from, you know, an object store somewhere all the way down into a GPU to do some work, and then all the way back out and sort of in a very efficient way, I think is interesting. And then on the product engineering side, you know, this is a very highly regulated space where you really have to worry about, you know, very keen on reproducibility and having their systems be modular and very well tested so that we can build systems that we have really high conferences working with, it's not just Oh, someone gets a 404. It's a problem like this is sort of impacting patient health. So it's really important that we're designing systems that are super reliable, super-resilient, and have really high uptime. That's pretty important. And then some other really cool initiatives, things like just how do you engage the pathology network? How do you build a slide viewer so that they can go in and annotate regions of a slide and say, Hey, this is what we think this is. So we can feed that back into the model training process. I mean, there's tons of engineering that goes around just like managing redesigning Photoshop, but in a low latency way, and it is tied into sort of a machine learning platform. It's a very challenging thing to do as well. And last, but not least, we recently, I think, last year, purchased our own physical lab. So we have our own lab space now where they do sort of all the lab woodwork. So how do you integrate this huge tech stack that we have all the way into these lab systems? where no one's really doing that yet? So we're sort of exploring this new space, figuring out what does it mean to get slides digitized and ingest them into this really large system and catalog them and then give that result back from our models all the way back to our own in house lab, sort of how do you build systems that do that kind of stuff is really interesting as well, yeah.
Details on the tech stack
At PathAI, we love Python, Python is great. It's not just a scripting language anymore. You know, it's, it's something you can use in production. So we're a big python shop. And to go with sort of the theme of reproducibility and modularity, we also love containerization. So Python plus Docker is a big workhorse of kind of what we get done here at PathAI. You know, the natural extension is okay, you're in Python and containerized, we, how do you run all these containers, we're big in Kubernetes. We love the Kubernetes orchestration system, we use it for both, like running our machine learning workflows, as well as a lot of our applications that we use the sort of back ends, or even sometimes databases and stuff like that are all based in Kubernetes. So not a requirement. I didn't know anything about Kubernetes when I got here, and so there's plenty of opportunity to learn here. As far as our front end work, you know, we like Typescript and Vue js, just for the type safety and stuff like that back ends, pretty classic stuff. Something like Django flask recently moving into fast API, because it's kind of the new hotness, which is really interesting, very occasionally, will drop out of Python. And it's something a little bit lower level, like rust or C++, if you need something highly optimized, which usually is only the case in some of our machine learning pipelines and stuff like that, but it's not uncommon. And our whole stack lives on a balance between AWS hosted services and our own in-house data center, you know, database, we use Eks for our Kubernetes clusters, or s3 for object store or RDS for databases. And then our data center ourselves, we have our own managed Kubernetes cluster, whereas I mentioned before, you know, we have 700 Something GPUs, we just sort of get cranking on all these different machine learning models. So that's kind of like the top down of our entire stack for the most part, yeah.
What to expect during the interview process
Yeah, the interesting process is, from sort of a interviewee's perspective, pretty standard, it starts with sort of a technical phone screen where we try and emphasize sort of what's your career PathAI? What are some projects are proud of, you know, what are some challenges maybe you've tackled along your career? And ultimately, what are you looking for? Why are you interested in PathAI? Which I think is kind of a great first taste of how do you get a sense of the culture, kind of what we like to prioritize? And then after that, it's very standard, you know, three technical questions, something like, you know, system design, and maybe two or two coding questions, and then finally a managerial question or interview panel. But we're really not interested in asking the types of problems where, you know, you have to write a depth-first search by from scratch or like, do you know that one cute trick to solve this problem? Just don't We don't think they're very good measures of, of how people work. So the questions are normally a little bit more collaborative, you know, you build up to making something. There are no wrong answers. It's more about like, what's your process for how you tackle these types of questions. And that's what really convinced me to join PathAI, because I was very intimidated from a lot of the other companies interviewing process where it was due this coding question and total silence. You didn't get it right. Sorry. It's much more of a collaborative experience. And that's what really floored me here. Because you could see in the interview process, I see it every day here. And so that's kind of what you should, uh, you should expect low, low stress, kind of just working on a problem together, you know, we'll get to some solution. And that's fine. Yeah.
Why now is the ideal time to join
Now's a really interesting time for PathAI, right, we're dealing with some of the most interesting problems facing machine learning startups, right, we have this really unique problem space where we're trying to balance development versus this for these highly specialized models, again, sort of being very rigorous and methodical, because the products we make aren't just, you know, about user retention, or about sales, like these are impacting real people's lives. And so it's the needle that we're trying to move here is has a really measurable impact. So that's been true, since it started. And it continues to be true, which is great. But right now, especially as you know, we're really focusing on growth and the impact that we have, which means a lot of the initiatives that we're working on are almost like making sure we're designing systems that are super sustainable and scalable from an architectural perspective. And we have lots of different initiatives. On the goes, you heard, you know, data center stuff, we have a lab integration work, we have contributed network stuff we have in house working with ml engineers, as an ML ops person like myself, lots of different projects, and we're, you know, we're just really excited about making those come to fruition and, and now's the time, we're sort of, we're moving out of, you know, the duct tape the startup life and moving into, like, how do we make this thing really sustainable and I think that's kind of a very interesting inflection point and a startup company. And on top of that, we just have a crazy good community of people, you know, people working together that have been working together for a long time now and some people new joining an already having big impact on the company. There's, there's not a lot of ego here. It's a lot of humility, a lot of transparency, a lot of collaboration is sort of some of our core values. So you know, we're just looking for, you know, motivated enthusiastic people to just come help build the future of high powered mythology, right. That's what we're interested in and I think now's a really good time to to carry the steam that we have, you know, for
Transcribed by https://otter.ai
