Listen on your favorite app
 |
 |
 |
 |
 |
 |
Welcome to Travel Buddy
In this episode of Travel Buddy, host Ian Andersen and guest Ravneet Ghuman, Head of Data Science and Machine Learning at Switchfly. Ravneet shares his journey from data engineering to data science and discusses the pivotal role of AI and machine learning in the travel and loyalty industry. He explains the differences between data engineering and data science, emphasizing the integration of AI in our daily lives. They explore the history and future of AI, including the significance of generative AI and machine learning. The conversation also touches on AI's applications at Switchfly, like enhancing user experiences and operational efficiencies, while stressing the importance of data privacy and explainable AI (XAI). The episode highlights how AI innovations, such as personalized travel recommendations, are reshaping the industry and what the future may hold.
Transcript
Brandon Giella (00:01.473)
Hello and welcome back to another episode of Travel Buddy presented by SwitchFly. Today we have Ian Anderson on the call as usual. Welcome back Ian. But we also have a new guest, Ravneet Ghuman, who is the head of data science and machine learning for SwitchFly. And he’s going to talk to us a lot about how AI and machine learning is used in the travel and loyalty space. And so I’m so excited to dive into that because it’s so important and such a hot topic.
So first, Rob Neid, first of all, welcome to the show. But second of all, tell us a little bit about your background. How did you get into data science and machine learning? Did you go to school for this? And then how did you get into it with Switchfly? And what are you guys working on there? That’s kind how I want to structure a little bit of this call at the beginning. So I’ll let you go from
Ravneet (00:53.006)
Sure. First of all, Brandon, thanks for having me. I’ve been in this industry for about 17, 18 years now. I started my career as a data engineer right out of college or engineering program. And I worked in data engineering for almost eight years. And then having been on the data engineering side, I decided
change gears or change careers and switch to data science where I pursued a masters from UC Berkeley and switched careers. And for almost last eight or nine years, I have been in data science, machine learning space. I joined Switchfly a couple of years back and I think travel is a personal passion of mine.
getting to work in a travel industry and apply AI or data science, machine learning to data related to travel, that really excites me and motivates me to create some interesting or solve interesting problems.
Brandon Giella (02:10.281)
Very cool.
Ian (02:10.681)
So I’ve known, I’ve known Revne for a couple of years now. Something that is immediately apparent when you meet him is, wow, this person’s way smarter than I am. he’s just one of those people you meet that’s, you know, he’s talking way up here. For us lay people, Revne, what
Brandon Giella (02:29.288)
Ha ha ha.
Ian (02:34.045)
Can you kind of explain the differences between data engineering and data science? Like what do they entail and how do you apply them?
Ravneet (02:45.07)
Sure, Ian is just being very humble. I enjoy our conversations with Ian. I think I see data engineering as sort of the plumbing needed to get water flowing from different pipes to where it needs to go. And being on the other side, data science is more about
using that water to create something like a cooking dish or sprinkling it into grass to grow something or lawn to grow grass, something of that sort. Data science is a more generic term that encompasses AI as one of the tools that data scientists can use to solve problems.
and it’s about predicting, forecasting, drawing inferences from data. That’s how I see the two sides, even though different, but they play very well together and actually they need to work together to solve problems.
Ian (04:02.546)
So I think, you know, when most of us think about AI these days, it still feels really new, right? The chat GPTs and co -pilots and all that stuff still feels really new. I mean, while the systems may not be particularly old, the kind of science behind it goes back quite a ways, right, Reveney?
Ravneet (04:28.11)
That is right. In fact, mean, the whole AI wave probably started way more than 50 years back with there being a couple of AI ventures wherein people had a lot of excitement about AI, lot of money was poured in, in the 60s, 70s, 80s, and then followed by ventures where people thought this is not happening.
What’s exciting and interesting now is since the big data revolution started, a lot of data has become available. At the same time, compute has become cheaper and faster. So now we have these old algorithms. Of course, there have been advancement in those algorithms, but we had these old algorithms dating back decades, new data, lots of data.
lots of compute and that was sort of a stepping stone in sparking this AI interest over the last 10, 15 years. there have been some advancements in algorithms. mean, what chat GPT is based on is, or type models are based on is called transformers. It’s a neural network architecture. And there was a paper written
like five or seven years back, that is what has led to Chad GPT. So even if I look at Chad GPT, it seems like it happened two years back, but the way industry has been progressing last 15 years, that seems like a very exciting and a natural progression, the way things have happened over the last few
Ian (06:17.968)
So it’s really the technology is catching up with the theory kind of is
Ravneet (06:22.894)
I think that’s a great way to put it. I think right now the buzzword is gen AI or generative AI. I think people use it synonymous to AI now and also to charge GPT type models. generative models essentially mean that you have figured out what the data distribution looks like
you can randomly sample data from there, from that distribution, or how the data exists. And you can start using that to make some intelligent predictions. So there have been generated models before. I personally think what made LLMs or chat GPT type models exciting is the previous generated models predicted numbers. So it was harder to sort of make sense of it, but text.
is very intuitive and you can understand. So now when you transform those numbers to text, makes things look exciting like a poetry written by a model versus a random set of numbers generated that don’t have any meaning to a person.
Ian (07:42.865)
And then, kind of on that, where does machine learning come in? And how does that kind of dovetail into the AI craze?
Ravneet (07:52.11)
So AI is sort of a higher level term that means a system that is intelligent enough to react, reason, respond, adapt. But machine learning is sort of a branch within AI that’s more about learning patterns. It’s more about pattern detection. So the idea being given you have enough data about
certain events that happen, you can apply some math and statistics to it and extract those patterns so you can start predicting future events that have not happened yet. Deep learning is sort of another niche that goes within machine learning and that’s more about applying neural networks or brain -inspired architectures that can help solve more complex problems like language.
or images and transformers, what I mentioned earlier is a type of a deep learning or neural network architecture. Deep learning networks are very resource intensive, both in terms of the data they need and the amount of data as well as the compute. I mean, if you talk about a model like Chai GPT, it
probably been trained on, I don’t know, all the websites in the world, the internet archive and whatnot. in compressed format, maybe gigabytes of text data and probably the amount of compute needed to train such a model is electricity needed by a small town. That’s how much power it needs. So deep learning networks are very resource heavy, but that’s an advancement.
of machine learning.
Ian (09:54.088)
Well, just shows how connected our computing and our data science is with other technology. I’d never thought about it being so resource demanding that now our power grid’s involved and transportation and all that stuff, how it’s all kind of interwoven.
Ravneet (10:22.222)
It’s interesting you make that comment. I heard someone else make a similar comment too in that when people think of AI, they don’t immediately relate to all the other things that need to happen to make it work. It’s just the technology. No, it’s not just the tech, but getting all the data. The moment you talk about data, then you need to figure out privacy
Ian (10:39.592)
right?
Ravneet (10:50.68)
mean, there was a controversy around a company using somebody else’s data to train a model. So then those things come into play. You need power. So where do you set up your data center or how do you train your models? You need chips or GPUs or specialized chips to train models. So is there enough supply for that? So I think maybe I’m not.
describing or covering even all the things or even half, but so many things play into
Ian (11:22.375)
Sure.
Ian (11:26.706)
Yeah, I bet. I think the very first time I remember hearing the term machine learning was these guys trained a computer to play Super Mario Brothers, you know, and it would...
the Mario would run and then run into the Goomba and then the next time he’d run and then jump right then but then fall into the pit and then, you know, and it just learned it over time. That was pretty incredible.
Ravneet (11:58.35)
Have you guys seen the movie around Google Go or how Google beat one of the masters in Go, Game of Go in Korea?
Brandon Giella (12:09.912)
No, I’ve heard a lot about it, but I haven’t seen
Ian (12:13.159)
Yeah, same.
Ravneet (12:14.638)
I think it’s pretty amazing to watch. Of course, there might be some dramatized components, it is it shows how I mean, basically the idea there is in a game of chess, it’s a 8 by 8 square. So there are a lot of moves possible at every instance, but still limited. But go being a much larger board, it just creates a lot.
of possibilities that it’s difficult for any machine to handle. So the way they created the algorithm and make it work, that was an incredible feat. Something that seemed impossible a decade back. And they ended up beating, or that game ended up beating one of the grandmasters of that game.
Ian (13:07.144)
Wow. That’s, go ahead, Brandon. Go
Brandon Giella (13:08.302)
I think, I was just gonna say that. I think it’s so amazing that you said that didn’t seem possible just a decade ago. Like a decade is not that long ago. Like a decade is like, was in, I mean I was in grad school, but like Twitter was just becoming popular a decade ago, you know? And it’s like, and now we’re creating all these tools that were.
to your point, like it’s impacting the grid. I mean, we’re thinking about how can I create a mini nuclear reactor for power a data center? It’s like, how do we get here in just 10 years? It’s amazing. Yeah, it’s just amazing.
Ravneet (13:45.274)
You’re right, like how the progress keeps compressing or the timeline keeps compressing. I mean, I remember 10, 15 years when people used to talk about natural language processing or any text -based AI. People used to create features with hand like, let’s extract these words or let’s do something, create rules. But now you have these new models. don’t need to create those.
units of features, you can have an architecture do that for you. Like that progress is incredible.
Ian (14:23.088)
that is wild. So where do you see, kind of, what do you see as the future of AI, both kind of in the next couple years, short, medium term, and then what is kind of the long term future of this look like?
Ravneet (14:38.798)
I think generally speaking, I think more and more integrated into our daily lives. mean, right now, we don’t have AI in certain places and when it is introduced, we see it as cool and amazing, but maybe fast forward three years, five years, we wouldn’t even think twice that we are using AI. So I think intelligent assistance,
helping us with our day -to -day life, navigating things we do every day.
I think that would be just way more integrated in a few years. I think, of course, there are those bigger and more tougher problems like autonomous driving, having a humanoid robot. I think we just have to wait and watch how that unfolds over the next few years.
Ian (15:41.907)
That’s really fascinating. you know, think like 30 years ago when they’re just kind of starting to stick computer chips and things to help, you know, like cars to help with diagnostics and things like that. And now we don’t even think about, you know, your car having a computer chip. Of course it does. Right. And
Yeah, I wonder if, you know, and this is just kind of the next evolution of that, right, is now it’ll be even more capable than it was before.
Brandon Giella (16:21.961)
I agree.
Ravneet (16:22.254)
Exactly. Everything has a chip now almost or intelligent devices. And I mean, there are pros and cons. I mean, one pro definitely is earlier car manufacturers needed to install certain hardware to enable some features. But now cars are so software and chip driven, new features can be rolled out like how
company’s rollout features on our cell phones. That to me is fascinating, something that was not possible even five, seven years
Ian (16:53.584)
like a computer update basically.
Ian (17:05.875)
Wow. And so we were talking before we got on, you know, we all have kids, very young. Mine are a little older than your guys’ kids. even so, for someone like me, I’m an elder millennial, right at the 40 -year mark. so I mean, I...
I have clear memories of not having the internet, for my adult life I’ve always had it. But our kids coming up in this AI world, I mean just looking at my kids with technology, it just comes so intuitively to them it seems. I wonder how AI is gonna be so much more involved in their lives than...
we can even begin to imagine.
Brandon Giella (18:04.442)
I’ve resolved that I won’t be able to help my daughter figure out what to do in school or for job or whatever, because there’s just, there’s no way to know at this point. I have no idea right now. But I do think there are, I’m a big humanities buff as Ian, you’re a history buff, but there’s some things in human nature, history, whatever that will never change and I wanna, you know, hone in on those school -wise, but as far as a job and like technical careers and whatever, I have no idea.
Ian (18:29.962)
Right. Yeah, I mean, they’re teaching they’re teaching basic coding, you know, at my kid’s school. And I think that’s because it’s just it’s not going to be a nice to have like it was for, you know, people my age. It’s going to be mandatory.
Ravneet (18:31.871)
I mean
Ravneet (18:49.198)
There’s so many job profiles that didn’t exist a while back. Influencers, mean, who would have thought 15 years back that that would be a new job profile. I think it’s still early in the AI game. It’s hard to foresee how things will change so quickly in the next 5 -10 years. Most of the larger companies are working towards AGI or artificial general intelligence
Ian (19:06.803)
sure.
Ravneet (19:19.566)
I mean, if there is a breakthrough in that space, which most experts predict in the next five to 10 years, we would see some form of AGI. I mean, the world beyond that, I can’t even imagine what it would look
Ian (19:37.543)
Yeah, that’s wild. So as Brandon mentioned, I’m a big history geek. I got a master’s in it. And I think about the internal combustion engine a lot of in just a couple of decades, like a really short time, we went from having a train, which was incredible and impressive and cutting edge technology. But then all of sudden,
Ravneet (20:06.915)
Woo!
Ian (20:07.568)
we can get rid of horses and they could fly and they could do things that they never thought possible all because of one technological change. And I wonder if this is going to be something like that of we have no idea. Maybe people like Ravneet have some idea, but people like me have no idea where it’s going and what’s coming. That’s awesome. So to bring it back around, Ravneet, what
Brandon Giella (20:26.878)
No clue.
Ian (20:36.471)
So we talked about travel a little bit at the beginning. We’re all passionate travelers and at Switchfly, you know, that’s our whole business. How are you using AI and machine learning at Switchfly? What kind of things are you working on?
Ravneet (20:57.799)
I think I like to see applications in two groups, like front of the house and back of the house. So we have use cases back of the house that support operations like consolidating hotel data from various suppliers using AI to help with some of those problems or looking at customer experience or how our service representatives are doing.
So there are some use cases back of the house and there are some use cases front of the house, like from the moment a user comes to a site till the time they make a booking, how can we personalize and improve their overall user experience? So starting from recommending destinations, hotels, to sort of guiding them through their booking journey.
I mean, we have done a lot of work there, but there’s a lot more that we can do. And especially with large language models, there is scope to sort of create personalized agents for every single person. So a lot more to come, but there’s some work that’s already been done in that space.
Ian (22:18.203)
Something that comes up a lot whenever you talk about data and personalization, it comes up in the media, comes up in legislature, is privacy. How is that factored in? I’m a marketer, I know just how much information I can gather on particular people based on buying habits.
demographic information and things like that. Where does privacy fit in? How are people kind of being protected against having data stolen and having too much known?
Ravneet (23:01.416)
I think privacy is super important and from AI point of view, think first of all, it’s important for a business to disclose what data would be collected on an individual, follow through some of the GDPR, CCPA regulations, whatever the policymakers have created. And those are pretty exhaustive and
sort of parameters that have been created in terms of what companies can store, what can companies store sort of life cycle of data. But I think still there is a lot more that can be done in terms of disclosing, first of all disclosing what a company can or cannot do. And then once that data is available, making sure there
there’s enough governance within the company to not misuse that data. And I think a big part of that is transparency. So also when AI plays a role in making a recommendation, it’s important to disclose to the user how that AI made that recommendation. So not just saying, we think you should go to Vegas, but why? Based on your history, we have seen
Let’s say you have checked out Vegas so many times or done so many searches, you have clicked on these hotels, there are new deals, we would recommend Vegas. So having some sort of explanation that sort of, I think an explanation builds a lot of trust and credibility for the company. And there was a time where companies didn’t have explainability in their AI models.
Ian (24:56.538)
You
Ravneet (24:57.698)
But in the last few years, I think that has really gone further and more companies are investing in XAI and explaining to the users why something was recommended.
Ian (25:14.502)
So what is XAI? Can you expand on that a little bit?
Ravneet (25:18.798)
XAI has become a term, explainable AI. Especially when we talk about deep learning models, it’s like millions or billions or trillions of parameters or numbers. It’s hard to know why model made a certain recommendation. Even if we talk about a model like ChartGPT,
I don’t think anyone can explain why it recommended something that it did. So there is this whole field that is evolving and a lot more work to be done. But that field is basically evolving to sort of answer some of those questions why something was recommended. And as algorithms are becoming more complex, XAI or explainable AI also needs to grow to sort of kind
catch up with the complexity of algorithms. But the basic idea is to explain why something was predicted or recommended.
Ian (26:25.333)
Wow, that’s fascinating. mean, it’s it’s so many, it’s a Pandora’s box, right? Like you open up this kind of AI thing and it just 90 other things kind of jump out that you got to do to catch up with it. That’s, that’s pretty incredible.
Brandon Giella (26:44.55)
I agree. I love the idea of for travel, finding things exactly like according to my preferences, but some things I don’t want that. Like I don’t want the same Spotify playlist over and over again, you know? So I’m excited to like see how it can explore more of my like taste, but also give me a little bit of excitement and adventure in my travel selections. Well.
Ravneet, I’m so glad that you were able to join us on this episode and share so much of your wisdom. Ian, I understand what you’re saying now. It’s like talking to Ravneet. It’s like, phew, right over my head. And I feel really dumb. But thank you so much for so much of your expertise and helping us understand AI a little bit better. I know it’s been such a huge topic over the last few years, but now people are starting to get a sense of what it can and can’t do and also explore a little bit of the possibilities.
Ian (27:21.606)
Yeah, yeah. Told you.
Brandon Giella (27:40.586)
I’m so thankful for your work and travel to be able to provide those kind of possibilities for us. So thank you for joining the show. Ian, thank you as always, and I’m excited to talk to you guys again.
Ian (27:50.152)
Thank you, Awesome. Thanks, Ravneet. Yeah, we definitely want to get you back on multiple times to keep us up to breast on what’s going on in the AI world.
Ravneet (27:52.214)
Of course. Thank you for having
Brandon Giella (27:57.741)
Amen.
Brandon Giella (28:03.866)
Amen, amen.
Ravneet (28:04.43)
For sure. Thanks for keeping it engaging.
Ian (28:09.374)
Yeah. Awesome. Thanks guys.
Brandon Giella (28:10.33)
Thank you. All right, we’ll see y ’all.
Ravneet (28:10.638)
See you later,
