"The moment you land from Google to the website, your experience should not be compromised." - Niraj Aswani, CTO and Co-Founder of Klevu
Commerce Famous Podcast, episode 17: Niraj Aswani: natural language processing meets ecommerce
In this episode of the Commerce Famous Podcast, host Ben Marks welcomes Klevu's CTO and Co-Founder Niraj Aswani.
The two dive deeper into Niraj's journey from a strong academic foundation in natural language processing to confronting real-world challenges in ecommerce search functionality. They examine how Niraj's expertise has been instrumental in Klevu's development of advanced search technologies, making online shopping more intuitive and efficient for users. The conversation also covers the broader implications of integrating AI and machine learning into ecommerce platforms, focusing on how these innovations are transforming customer experiences, improving search accuracy, and influencing the future direction of the retail industry. Enjoy!
Listen to the episode right here or subscribe to Commerce Famous on Spotify, Apple Podcasts, or your preferred podcast player
On generative AI and ecommerce: "It's a continuous feedback that, if we give it to the machine, at the end of the day, it is going to make shoppers life easier. That's what matters." - Niraj Aswani
The best ecommerce and retail podcasts
Transcript of Commerce Famous episode 17, an interview with Niraj Aswani
Ben Marks: Hey, everyone, welcome to the commerce famous podcast. I'm your host, Ben Marks, and with me today is a technical co founder CTO of Klevu, Niraj Aswani. Now, Niraj got his start at University of Sheffield in natural language processing, so I expect a lot of, especially AI and machine learning insights. So, welcome to the Commerce Famous podcast.
Niraj Aswani: Thank you very much, Ben. It's my pleasure to be here. I've been hearing a lot about this podcast and it's great to be here. Thank you.
Ben Marks: Well, mostly good things, I hope. So far we haven't received too many complaints. So Klevu is a Finnish company, but you are at home in the UK, right?
Niraj Aswani: That's correct, yeah, that's correct.
Ben Marks: And just sort of reading through your background now, folks, Klevu has been around for eleven years now, right?
Niraj Aswani: Since 2013. Wow. Yeah.
Ben Marks: And if my reading is correct, you got a start on Klevu. I mean, you co founded, this was sort of your. You got your doctorate at University of Sheffield and you went out and basically co founded this company. And I would say everyone probably in this business knows Klevu's name and knows what you do. How did that come about? Right, because by the way, folks, I should not bury the lead here. I have with me on the podcast the author of designing a general framework for text alignment case studies with two south asian languages. So this was your thesis, right? This was your doctoral thesis in university. How did you get from there into.
Ben Marks: And now I think I know the answer, by the way, but how did you get from there into co founding Klevu?
Niraj Aswani: Yeah, Ben. So I actually came to the UK at the University of Sheffield to do my master's in natural language processing. I really loved that subject. And I had heard really good words about that university, specifically about the natural language processing group over there. So I went there, I did my masters, and then I got this wonderful opportunity to work on one framework. It's called general architecture for text processing. And as part of that project, I got this opportunity to work with many european projects which actually involved working with quite a few industries as well, industrial companies like BT, we work with Siemens, and we were also working with some government projects as well. We were working on some medical projects as well, where we had to process lots and lots of text and try to understand, try to make sense out of it.
Niraj Aswani: The wonderful thing, and the funny thing actually was that when we were dealing with the industry, we saw that industry was hardly using any of the NLP technologies. And there was a huge scope over there. Specifically, when we talk about the e commerce, it was most likely only Google who was understanding what customers were typing there. So that's where we actually saw the opportunity. Now, my colleague and our CEO, Nilai Oza, he was studying at the Hutfordshire University. He was doing his phd at the university. And after ten years, we actually got back together. I had my technical background and he's an amazing guy.
Niraj Aswani: He knows how to bring an idea from lab to the world stage. So we got together and we decided that whatever we will do, we will do it together. We will utilize our complementary skills. At that time, we did not know what we were going to do. And it started with me resigning from my job. And shortly after that, Neelai also resigned from his job. And then we started a company whereby we were fortunate enough to get a few NLP projects from his contacts. And that gave us some Runway to run that company for some time.
Niraj Aswani: But whatever we were building at that time, of course, with the customers consent, we were building some reusable components. And at that time, the AWS had come up with this idea that everything should be a web service. So that's how we started building components. We started building everything as a web service and eventually, at one stage, Nilai actually came up with this idea that Neelaj, we have so many this individual components, we have the NLp expertise here. How about we try to build some prototype for an Ecom website? We did that and surprisingly, it was working really well. The idea was that you have wonderful experience on Google, but the moment you land on any e commerce website, your experience was really poor. How can we improve that? So that's how we actually started focusing on the on site experience. The moment you land from Google to the website, your experience should not be compromised.
Niraj Aswani: It should remain the same. Yeah, so that's how we build our first prototype. Yeah, I think Nile was the one who reached out to the initial investors and, yeah, I think he secured the funding for us to take this.
Ben Marks: Mean. And I did kind of gloss over detail here. So this explains the name of the company that you were both briefly a part of. Ni squared. Ni dash squared.
Niraj Aswani: Right.
Ben Marks: Because of the similarity of the first names there. And I didn't realize that you all had known each other for so long, but I'd love to just focus on something you said, which maybe you all just knew you were that good or just youthful optimism, but you basically said the only company that was doing anything like what you were proposing to do was Google. Basically, I think you said, know what the customer is typing, right. And not just like sort of, we can look at the data after the fact, but this is also real time. Understand what's happening. And you're right that back in the day, back in the early two thousand and ten s, the commerce experience was still largely, I think, driven. It was very search driven. And it was driven.
Ben Marks: Your vector into any given website was via Google. Right. And so it's actually kind of genius building this sort of connective tissue where the site has both context and continuity for what the consumer was doing before they even hit the site. That's a fascinating insight into doing this.
Female Narrator: Commerce famous is proudly presented by Shopware, the leading open source ecommerce platform for midmarket and lower enterprise merchants. More than 50,000 clients already process over $25 billion in annual GMV through Shopware. Find out more about Shopware and the best value in ecommerce@shopware.com.
Ben Marks: I actually went and I took a look. I went to thearchive.org, which everyone listening should support. It's an invaluable tool, preserving the history of the Internet. But I was looking at the, in 2014, I figured I'd give you all a year. So I think it was like March of 2014. There's a scrolling image, hero banner, and it was the smartest search box for the online store. That was your tagline back then, right? But when you get down into it, I remember when you all started showing up at the conferences, like at the, you know, at the ecommerce conferences, and this was nascent and it was aspirational. But you could see the logic already because what we knew at the time was that the search facility that existed in basically every platform absolutely sucked.
Ben Marks: I mean, a lot of them wouldn't even handle a typo well. But if there were homophones or if there were just anything off, the results could either not show up at all or they would vary wildly. And I think you all were aware that, hey, there's this trove of data out there. And just if we partner with these platforms and if we get ourselves installed across multiple, big enough footprint of websites, then we can leverage our computer science skills, our NLP skills to actually deliver meaningful results. Now, was this all just apparent right from the beginning, or did this just kind of come in iterations?
Niraj Aswani: Ben, you highlighted a really well point there. So if you look at any platform, the platforms are, I mean, e commerce platforms, they have so many things to are. When we evaluate any platform, Shopware, Magento, Shopify, you name it, they have so many things to look after. That's where we actually saw the opportunity. That search is something that's not been looked after really well at that time. We are talking about 2014 here, and we actually came across one really good research paper at that time, which talked about the analysis of top 50 us retailers. And they actually highlighted the types of the queries that shoppers were actually firing. Oh, wow.
Niraj Aswani: On this website. And they highlighted twelve different types of queries that customers were firing. And next to it, they highlighted. What was the support provided by this different platforms. Actually, the most of them supported the keyword search, which is matching exactly what you are typing there with. If you find that in a product, that's what they will show over there. But there were other types of queries. I'll talk about some of the complex ones.
Niraj Aswani: So one of them was symptomatic search. Now, what it simply means is that if you have a headache, you don't know which medicine you want to take, but you go on a website and then you start searching, like medicines for headache. Hardly any platform would support that. And then there were thematic searches, like somebody searching for, I have a big lounge, the walls are brown or whatever, the light gray, and which so far is good for my lounge and system wouldn't understand. So even back in 2014, people were searching with queries like that, but none of the platforms at that time supported that. So that's how we actually saw that opportunity. That if we can only focus on that one aspect. And by the way, some research also suggests that a human brain can remember anything between 10,000 words at a time.
Niraj Aswani: And the challenge that we saw for the merchants was to create their catalogs that has words from this, between 10,000.
Ben Marks: Words from this vocabulary.
Niraj Aswani: There are so many shoppers, so many different shoppers, so many different variety of shoppers coming on your website. How on earth a merchant is going to write a catalog that understands every single kivia? So that was like a booster for us, that, hey, there is a significant opportunity here. How about we automate everything? How about we read the mind of a shopper and doesn't matter what they are typing, let's connect it with what's there in the catalog. And that's how we started, Ben. It's been over ten years still. Every single day we receive a use case from a merchant that we feel that yes, this is a new thing and we should be working on that.
Ben Marks: So the only tragedy behind, well, one of the only tragedies behind this being a podcast is what people can't see, which is what I see right now, is that there's a genuine enjoyment, satisfaction, and I think fascination still evident on your face as we talk about this. I think it's coming across just fine in the audio, and I love that this is far from being, quote, a solved space, much like ecommerce. I don't think we're going to solve this problem unilaterally. Put a pin in, it's done, we won't be there anytime soon. And actually, for me, as a bit of a segue into, I think it's maybe a little bit, I don't know if it was prescient or possibly maybe just a little bit lucky that you all working in a space, I mean, NLP, if you look up natural language processing, it's a pretty good window into, I think, what it is to be a human being. And as such, as we as a species really embark headlong post Chat GPT, or GPT four, where the rest of the world has awoken to the possibilities of AI. Working in NLP for over ten years, that gives you such a head start, because fundamental to that discipline are machine learning. Now, I guess if I'm being a responsible host here, I have to ask, so I don't want to gloss over something I've said in the past, which is that companies, especially companies around recommendation personalization, like these companies, have been dealing with large data sets and have been dealing in the ML and AI spaces long before it was as hot and appealing as it is today.
Ben Marks: But I have to ask, after Chat GPT came out, right, I know it had an immediate impact on the strategy of shopware, and it had plenty of impact across plenty of industries. Did that have a significant influence in your business or was it just like, okay, great, now the rest of the world is waking up to what we've been doing.
Niraj Aswani: Yeah, so large language models have been there for a long time, but the way OpenAI actually productized it and they converted that into a chat JPT, they have actually made it available to every single person. They have made it possible for the persons, to anybody to ask anything they like. And surprisingly, it works amazingly well. Of course there are some downsides there will be certain concerns around it. For example, the one that's most talked about is hallucination happening. The other one is what happens to your personal data? Are you submitting any personal information to it? Then what happens? How these companies are utilizing it. So just one thing to mention there is that OpenAI is not the only company who has large language models made available. There are so many companies.
Niraj Aswani: Google's Bard is there. Facebook meta platform has its own large language model. There are so many open source platforms available, large language models are available there. But I have to say this, that the way they productized this entire large language model is something worthy of mention in every podcast that we do. How much impact has it? I think it's a happy problem in a way that people have started understanding that what are the capabilities of AI, how AI can actually help there. I remember earlier we had to explain to individuals, merchants that what is the meaning of a semantic search? What do we mean by that? I think today if we just say like the way the Chat GPT is working, the way intelligence has become available as a service, that's what we have been doing for a long time now. And you can get the similar experience if you're using our product.
Ben Marks: Okay, well, I guess the next question is you brought up a good point, which is that especially in the early days, your ecommerce platform was the hub. It was almost a nerve center for almost everything. And whereas that was the case, ecommerce platforms tended to become, again I'm speaking especially early 2010s, tended to become jack of all trades, master of none, right? I mean, maybe you still had good catalog, good cataloging and stuff, and some of the more rote or commerce functionality was there. But anything outside of that, even the content management aspect could suffer from not being purely focused on that one mean. I think this is part of the theory behind for enterprise businesses, for the mock implementation pattern. I know you all are part of the mock alliance, and certainly for businesses of size, that can be a right approach, I think, right, because you have just the large enough scope and frankly budget to be able to value that flexibility and composability against the cost of building it out and maintaining it. But with AI coming in, there is this emerging. I thought that the machines effectively can become the orchestration layer.
Ben Marks: They can sort of take a lot of the burden out of what is now a fairly expensive process, at least to engineer in the beginning. And I think the same sentiment could even push down into the mid market. And then we see adjacent to this what the big aggregators in the room. Amazon, Facebook, Shopify, they will all use this in their own, well, I should say will use this. They are using this in their own way. But do you see a future where Clavoo sort of becomes irrelevant or consumed by some MLAI apparatus that just sort of knows how commerce is supposed to work?
Niraj Aswani: Yeah, so I see it this way, that there are two types of, well, actually there are three types of opinion in the market about this AI or generative AI, I would say. One is people who are really scared about it. They see like they think, this is not for us, let's not use it, we cannot trust it. Then there are others who are really excited about it and they want to do everything using AI, but they are unsure of how to control what to give it and what not to give it. There's a real thread there that these tools have to be utilized ethically, I would say. And it's the responsibility of individual companies to make sure that whenever they are making these services offered by these tools available to any shopper, they make sure that the trust of the shopper is not broken. And that's where I feel that it's a responsibility of a company like ours to become an interface in between and make sure that whatever shopper is submitting to us, not everything, we are submitting to this AI tool.
Ben Marks: Right.
Niraj Aswani: We are building this filters in between to make sure that only the relevant information that is non personalized, which cannot be classified as a personal information, together with the information which has the absolute answer of what the customer is asking for, that is being submitted to the tools like OpenAI, and the responses are actually given back to the shopper, which promises no hallucination or the least possible hallucination in the responses. So, yeah, as a clever, I think I'm not afraid of this advancements coming up, but rather more excited about it because I know how we can utilize this technology at very best, ensuring that customer has the confidence that their information is safe. It's not going anywhere over there. Yeah, we have in fact, Ben, sorry, I'm not marketing anything, I'm not a salesperson.
Ben Marks: I wouldn't expect it from you.
Niraj Aswani: But I just want to say that we had been working on some of these tools for a long time, and now new products are actually coming out from Claywe sent, which are there to bridge the gap between the online and the offline shopping experience.
Ben Marks: Interesting.
Niraj Aswani: Everything that you can do in an offline shop that will be possible through some of the new tools which are coming to do on the online world. Yeah, I don't want to go much more into.
Ben Marks: I understand.
Niraj Aswani: Generative AI.
Ben Marks: You're allowed to be excited about this, right?
Niraj Aswani: Yeah. I just want to say that using generative AI, without generative AI, that was not possible earlier, and that's what generative AI is making possible. Yeah. I'll just give one example. Let's say you're looking at one particular, say a lamp, right? You're looking at a lamp, and you are not sure about which bulb will be required for this one and how long that bulb would last, or what is the cost to maintain this lamp itself, or like continuous basins, buying the same lamp or things like that. Those kind of questions you would usually ask when you are in an offline shop. The challenge is, can you make those kind of question answering that kind of conversation possible in the online world? And some of the things that is possible now because of the generative AI, I mean, all of these things are possible now using the generative AI. And that's what at Clavu, we are utilizing and bringing the next generation tools out.
Ben Marks: Okay, I think I'm in full, maybe a bit naive on my side, but I'm prepared to offer my naive agreement here, because generative AI is. It was a lot of people, and really, for me, it was the first time I saw, okay, I can now see how this thing is functioning, and it's doing something way beyond the capacity that I expected. But there, of course, was immediately the hype cycle on that through the roof, and that has this, I think, overall depressing effect on the technology. But then we start to see people in businesses incorporating generative in useful ways. And for sure, I agree that these kind of interactions that you describe, where you have this sort of blend, like, I wouldn't truly know whether I was speaking or chatting with AI, as the case may be. All of this goes to, I think it just goes to show that the claims of any platform offering AI based functionality should be evaluated on the merit of what they're actually producing and whether or not that moves the needle. And we could probably all just thank our forebears in this space. I would say clay was probably among them for sort of delivering this industry into this meaningfully into this era of AI.
Ben Marks: I'd like to bring up one last point here. Back in November, I read an article, I was in the New Yorker. It was by a guy named James Summers, and he's both coder and a journalist. And the title of the piece, I've been recommending it to people when the topic comes up is a coder considers the waning days of the craft, and he just basically takes on generative AI in the context of being a developer and his thesis or his hypothesis, as much as I think I could get at it, was that there will be an adjustment. This is an industrial revolution scale technology in terms of economic and employment impact, but that the skills that will serve people really well. Get back to something which I think you were just kind of implying, which is sort of like, hey, you've got to understand. You sort of have to incorporate a bit of the machine mentality. You almost have to empathize with the LLM or the perspective and the data set that are being consumed, so that you can actually almost partner with these tools and technologies.
Ben Marks: And if you're doing it right, and I think this is the point you were getting at, Niraj, if you're doing it right, then the tools themselves will improve. It's this sort of virtuous cycle of seem that's not too rosy a picture, right? That is what you expect to happen, isn't it?
Niraj Aswani: Absolutely, Ben, you are absolutely correct. We have to work together. Even this large language models how they have been developed. That's based on the feedback that humans have given to them. And we have to keep continue doing that for different domains as well. Like in shopping, for example. Right. Today, we may be given a question, we may be producing certain answer that has come from the catalog, but even after that, customer may ask questions for which there is no information available in the catalog.
Niraj Aswani: So there also, merchants would have to work together with the machine. They would have to train the machine. That if a question like this is coming, this kind of answer should be given over there. That's a continuous feedback that if we give it to the machine, at the end of the day, it is going to make shoppers life easier. That's what matters. That is what matters at the end of the day, that shoppers should have that happy moment when they are shopping. It should not be delivery is not the last thing, but once they open it and if they find it, that the product is as it was promised, as it was expected from them, that will give that happy moment to the shopper. And using the generative AI technologies, I feel it is possible now, even in the online world, that you can establish such an interface.
Niraj Aswani: You can build such an interface where customer has the opportunity to resolve any question they have about the product, things which are there in the catalog, things which may not be in the catalog. Using large language models, you can utilize that general knowledge that's available in the large language models. But it is possible, it is possible that you can satisfy all the queries of the customer and give them that happy moment that we just talked about.
Ben Marks: Well, I think that's a perfect note to leave it on because I'm a huge fan of stewardship in this space, and I think as long as the vendors in ecommerce are focused on the stewardship of the end customer, that's a guaranteed recipe for success. Now, before we close things out, Nirash, I just want to say I should also thank you was Rachel had me on your discovered podcast.
Niraj Aswani: Discovered.
Ben Marks: And I was on there with a good friend of mine, Mayer Beemire agency up in New York, and we had a great discussion. In fact, that episode just came out. So that was a real pleasure for me. And it was actually great to actually talk about some of these related topics here. So, folks, feel free to go over there and check out that podcast as well. But for now, Neeraj really can't thank you enough for taking some time with me in your afternoon to be on the commerce famous podcast. And I'm looking so forward to the innovations to see what you all have coming up in the near future.
Niraj Aswani: Thank you. Thank you very much, Brent, for having me here and listening to me. I think listening is also an art and you do this really well. Thank you. Yeah. Would love to hear feedback from whoever is listening this. You are actually part of our journey as well. So if there is any feedback, any feature that you would like clever to work on, I would very much welcome your contribution in our journey.
