Build AI Apps with (Customer) Context | Webinar July 2025

0:05 All right. Hi everybody. Thank you for joining today. My name is Teao Gonzalez and you'll meet Quinton Wall here shortly. we're excited to be hosting today's webinar about customer context and AI applications. we're going to give people maybe another minute or two before we go ahead and get started. some housekeeping items. Feel free to leave questions as they come up in the chat. either I or another member of the team who's participating will be able to to answer those as they go.

0:36 that said, thanks for joining today and looking forward to getting started soon. Looks like we have a bunch of people joining chat, which is great. Yeah. All right. And we got another tail over here typing. So, it'll be a good time. First tile. One of those.

1:14 Nice. Thanks for joining us over here. I'm going to assume it's the same pronunciation, but in case I I'll be doing the Teoto as we go here. Okay, we're gonna start in 10 10 seconds. Quinton, hit the New Year's Eve clock, you know. All right. I wish I had a little animation to count it down. Yeah.

1:46 Yeah. No worries. Okay, cool. well, thank you everybody for joining. As I kind of mentioned, my name is Teao and so I'll be doing a bigger intro later, but as you know, you've joined us for our context and AI applications webinar. We're over at Airbet. we think of ourselves as the go-to data inestion platform from any source into any destination and and help you ex execute on analytics and AI use cases. But the focus of today is really just leaning into that AI part. so so what Quinton and I are going to be sharing more on is is just talking about what's going on in the market with AI applications. We want to talk about what trends people are seeing. and then what's what we're most excited about is actually to talk about the infrastructure that we've heard is what m that we've heard and been able to see and executing with our customers that makes up a really good generative AI application. And we're also going to talk about how you go ahead and provide context for that application. and we'll talk a little bit about a new product we have called Airbite embedded and we'll have time at the end for Q&A. But that being said, you can feel free to go ahead and put it in the chat, too. No need to wait until the end. that said, I I promise we do some intros. So, I'm I'm our head of AI business development here at Airbite.

3:02 but Quinton is the much cooler person. So, Quinton, how about you go? Hey, thank you for that. I decided to use AI to generate my profile picture, and now that I look at it on the webinar, it's a little bit creepy, but my name is Quinton. I run Devril here at Air Bites, being an app developer for a long time and been completely amazed at how quickly AI and the different concepts are coming into change how we build and develop apps. So super excited to talk to you all about that. Got it.

3:30 Thank you, Quinton. so with the next one, what we are going to go ahead and do is let me actually do let's talk about the current state of events. And by the way, if people can I get confirmation from someone in the chat? Are you able to see the slides in full screen or are you only able to see them in an equal kind of size with me and Quinton? I know I have my friend T over here to help me. Oh, and I have full screen. Okay, perfect. Just wanted to confirm. so let's talk about what's going on in landscape. I mean, you're here for this webinar. You obviously have an interest in AI. And the good news when you think about it and we can go to to the next screen and and take a look at some of the trends that make this really relevant. it's kind of booming everywhere, right? I mean if you work at a company or if you have your own company, you're probably thinking about what your AI strategy is and how you're incorporating it not only internally but also for your own customers. Quinton, we were just chatting about like the different dev tools that we're using that that really lean into generative AI. And I don't I don't know if maybe you had already known this Quincon, but I think at the beginning of the year, Chat GPT had 400 million users or 300 million. So when I looked at the stat last week, it's 800.

4:40 I'm just like everything is popping off. it's just one of those things that GBT is like the new Google. It's something that everybody uses every day, but that's the tip of the iceberg, I think, of what people know and think about when they talk about AI tools. It's kind of the search and question engine, but there is so much more that is happening out there, whether it's agents or idees or anything that's happening with inside there. It's a fascinating time. Totally. and and the stats speak for themselves. So, rather than even go into those, I think I'll just share anecdote towards Quinton's point. In the lunchroom the other day, a teammate was talking about how their their child doesn't even know what Google is. They actually ask all their questions in chat. So, that is the moment we are in right now. we shall see what happens. Gemini is making big moves. So, that being said, beyond the cool things going on in the market, I think one of the things that Quinton and I have been talking a lot about, and you feel free to kind of chime in if you have something you'd want to add here, but it's really interesting when you talk to people who are building on top of these large language models or any other AI infrastructure, just what is top of their mind. so we we spoke to like hundreds of people who are building things from enterprise search for their own organizations or they're building like voice agents for their own external customers. And the consistent theme that we've heard is beyond just everything that goes into building a great AI LM driven AI application. it comes down to having the right data and having that data processed in the right way so the LM can work its magic. And so we heard things like data is your remote all the way to picking what types of data you want to leverage is really important and that was consistent whether you were a large company that works in the physical good spaces like Lowe's or you are a company that's building out these foundational models like open AI or MRL. So data is super important here and I think yeah go for it. I was going to say it's interesting because obviously Airb was founded as a data movement platform but our one of our co-founders Michelle he has this quote that always sticks to me as well is like without data there is no AI. So, as a result, we've been in these conversations about like moving data and having context available for a long time, but just the people that are using it and the amount of data and how they're adding that like contextual information, it's it's it's supercharged over the last year, I would say, just in the sense that how important data is to these public LLMs, but doing it efficiently and governable so you don't share the wrong information. Totally.

7:19 Yeah. And you're going to hear us talk about context. We think data is like a huge part. H I'm so with it. Like data is king and queen. Data runs the show for sure. and I think context is the word we'll probably use more often when talking about it today because there's not just your plain data as is. There's a lot of things that you're going to do on top of that. So context is what we're going to lean into the most today. Is that fair to say, Quinton? Yeah. And I think that's really relevant because when you think about the way something like an LLM works that knows effectively the world's public knowledge, but it doesn't necessarily know your corporate knowledge or information or even your customers knowledge and information. So it's that contextual information that you provide to an LLM in a secure and governable way that I think is as that first quote says that's kind of your moat that's your unique differentiator that you want to be able to leverage as much as you can. Totally. Okay, let's lean into that a little bit further. So, let's go over to talk about some apps that we got the chance to kind of do deeper dives into talking members of the team. whether they they work with us or not, I think what what really stood out and two use cases that I wanted to talk about were more the Glean and Sierras of the world. So, Glean, if you haven't heard, they they're very popular for enterprise search. I think they were probably one of the main pioneers of this function, which is you have tons of documents all over the place. you need to ask a question of your enterprise and get access to the right document at a given point in time. they also release agents that kind of assist this function as well. But Glean has gone ahead and and built out infrastructure that largely gets you access to the context for your company throughout every level and surfaces what you need.

8:57 And Sierra is doing the same but more for their own customers customer support experience. So you've probably also heard of companies like Decagon. there's all sorts of people who are handling this customer support use case. And what's really interesting is whether you are an enterprise search customer and you're an individual human who's going ahead and asking a question or whether you're an AI agent that's trying to figure out what's the right response to a human inquiry. Maybe there's a issue going on that's not too sophisticated. the AI agent still needs to go ahead and grab the right context. maybe from Zenesk, maybe it's from SharePoint to be able to adequately answer the question that's been asked. And so these are two really great examples of companies that are crushing it in my opinion. and I think doing really cool things that a lot of other companies are looking to emulate as well. but Quinton, I think this is where it's interesting. It's a common architecture for both of these. Fair say yeah it's it's kind of a perfect kind of segue that we are really seeing a common architecture or blueprint for these apps and a very high level you can kind of think about it on the back end there's some type of data hub that allows you as a a customer allows your customers to easily on board and share their contextual information. So if you look at Glean or Sierra, they're providing a an app that pulls in their customers data and shares that in some sort of data hub where you can manage the credentials and govern what is shared and you move that to something like a staging store like a data lake or S3.

10:31 Once it's in that staging store, the app developers that are building these applications can then access and process that data by putting it into say a vector database or something similar where they can add embeddings and enrichment and document chunking if you're sending files across. So when they build their applications using whatever modern stack language that they want to use whether it's React or TypeScript or Node or or Electron they can build that leveraging the database and the contextual information in say PG vector or something similar and share that information with confidence to a public LLM like OpenAI or Gemini or whatever they decide to choose. So the customer can gain insights from all of that information. And this blueprint and a pattern is very very similar across many of the apps and customers that we talk to. Now, when you kind of double click down a little bit as well, what's really important, and I think we touched on this a little bit before, is it's not just about sharing all of this information in particularly a customer's information with a public LLM. This is where you need to be really careful about what you share. So data governance in that pipeline is still as critical as it's ever been, if not even more. So what you need to be able to do is when you extract that information from after the customers have onboarded and created those sources, you still need the ability to decide what data you want to share. Maybe you want to encrypt it or do some sort of PII masking or any of those areas and map it to where you need that data to go. So when it gets shared into a contextual store, you have confidence that the right data is there.

12:14 So once you start adding the embeddings or anything on top of that, the LLM will query as we know it does in say a rag application, but it's querying what you've provided and made sure that is publicly available to that LLM. So the app developers can start to use that information. Can we on this note Quinton I think I really like your point like there's so much more that's happening in this pipeline than I think what people typically think of inherently when you think of like ELT ETL paradigms like one of the examples we keep talking about with our customers is when you're building out your application you have your LM going ahead and maybe executing like a hybrid search of of your vector database or or whatever sets of data you have on whatever platform you're thinking about not only grabbing the content, but you're also grabbing all that other stuff. That's super important, right? Like you don't want your your application to go ahead and be sharing salary data with the wrong person in your company. Like that's a huge huge risk. And so you're thinking not only about that data, but those details. You talk about governance. Like that's something that we are probably going really into the weeds with people when it's like you need to be able to capture all of that data so that your LLM has the full picture and executes on the outcome you're looking to achieve.

13:30 So I think it's like a super good call out. Yeah. And then when you overlay things like files are now a huge part of the data and the context that's being served in that's unstructured data. You have no idea what that file structure looks like even in your own business. But when you start adding customers contextual information and their files things like iceberg and schema revolution are really key and fundamental to what we're seeing in the in the market space here. just so you can keep evolving not knowing what that structure of that data is actually going to be and then serve that into that context store. Yeah, that's a great call out. But I guess if you think about it, Tio, in the end, the challenges kind of remains the same if not accelerated.

14:13 Like getting that right customer context into the hands of your LLM is tricky and hard for all the reasons that we we just kind of mentioned. But who is actually using that context we're seeing evolve a lot as well. Traditionally, we've seen folks like data engineers working with data pipelines and integration processes, and that's kind of the bread and butter of where Air Bite has come from. But more and more now, we're seeing the need for these AI app developers that want to tap into that context store after all of that customer data has been fed in effectively so they can build those apps that I mentioned before in that blueprint. But we're also seeing the rise like everybody else is of agents and LLMs and robots that are consuming this data in context as well.

15:01 So I think we're really hitting just like a velocity point now where there's a finite number of data engineers and AI app developers that are super important. But as agents and LLMs and agentic workflows grow and grow, we're only going to see this accelerate for sure. Absolutely. and I think on that on with that I mean the current way we're servicing a lot of these use cases right now is everybody embedded but the reality is also everyone like when you're building your applications like you're making these very conscious decisions is like what best fits your stack what best fits your architecture the reason we went ahead and built embedded is because we realized as Quinton pointed out is expanding the the in the AI era we're expanding the scope of not just serving data engineers but it's those application developers and agents themselves and what that usually entails sales is thinking about a paradigm like Airbite of ingesting that data from source to a destination and and the pipeline in between at scale like you're doing this for hundreds or thousands or hundreds of thousands of customers because you want to make sure that your LLM is able to deliver quality experiences as you grow your business and as you grow your customer and user base. And so with everybody embedded a lot of the focus is on go for it Quinton. I was going to say maybe to step back a little bit on what actually embedded is. It's an implementation of that kind of AI data hub blueprint that I I showed you earlier because we're hearing that from customers needs. Yeah, totally. So like when you think of the AI data hub, I mean you're going to see it much more with Quinton. So we we won't bore you with like im with text and abstract concepts without you seeing it. But it's exactly that like bringing the data the data AI data hub straight into your application by giving you the integrations giving you the multi-tenency required to handle customers at scale allowing you to handle structured and unstructured data all while ingesting and managing that experience across a huge customer base.

17:01 so that being said we want it to be a fast experience. You should be able to do this pretty quick and that's what we aim to do with embedded. So Quinton maybe it's just off to you to to show the how it works. Yeah, I think I think we've done enough talking slides so far time. Let's see if we can live on the edge and do a live demo. So, I'm going to switch over to I'm inside Airbite.

17:21 So, if you've used to and familiar with Airbite, you've probably seen this set up before. I have my connections and sources and destinations on the side. But what you'll see here is I have an S3 bucket setup. This is effectively that staging store that I showed you with inside that blueprint as well. I won't go through all the details here. If you're used to S3, you're pretty familiar. But what we're doing is we're setting up a central staging store with inside that data hub using embedded.

17:52 What it means is once I start connecting customer information and their context, it's all going to be fed to a single store. But instead of me as say a data administrator or a data engineer getting in the middle of creating connections and managing credentials for all of my customers, I want to get out of that process and provide an easy way to allow customers to onboard themselves. And this is where embedded really takes off. So I'm going to jump over here. I've created just a small little React application here. And what this is simulating is an onboarding app for your customers. So let's say you're building an AI app using embedded for that data hub in that staging store. I now want to onboard all of my customers very easily using any of the 600 plus connectors that Airbite provides out of the box, not counting all of the ones that may be created by individual users.

18:45 But let's say and I'm just going to use a my an email here for a unique identifier, but this could be anything that your customer uses that you provide for them. So I'm just going to call this demo3 at women webinar.com. Don't worry, not not a real account. Hopefully somebody doesn't get that. Now once I've set up this kind of unique identifier embedded provides the ability to onboard customers via an API or better still via a widget which is what we're seeing here. Now these tiles that you see here as the administrator of that embedded environment or that data hub you can expose whichever credent whichever connectors that you want. I've just got a few set up here and you can decide on which ones to share with your customer based on what you want them to connect as their context. So I'm just going to go and create very quick one using Faker data source. Now that's that's going to go off and do its thing. behind the scenes. What it's actually going to do is it's going to create a separate workspace with inside my Airbite environment that I can manage and take advantage of everything that Airb as a platform provides. That one's still going on in the background.

20:01 Hopefully, it will show up. So, what you'll see down here is it's gone ahead and created a separate workspace with inside Airyte embedded to create that data hub. So you'll see I've got a bunch of different demos here. These are all individual customers that I've onboarded. If I click inside here, what it's gone to do on the back end, it's gone and created that source that the customers provided and they would provide the credentials. My simple example did not do that obviously and it's going to send all of that to that centrally configured staging store. So you as an administrator, you don't have to worry about onboarding any of your customers to do that. So it takes advantage of all that stuff really easily for you. Now how we've implemented it, what we've found is that most customers are very JavaScript framework heavy at the moment. So things like React and Node, TypeScript. So we've gone ahead and created a JavaScript library that implements that widget. And then inside my React application here, you can see all I'm doing is creating a very simple React component here to handle that connect data. And if I click straight into here, you'll see all it's doing is figuring out a way to open up the widget, provide a unique identifier.

21:18 In this instance, that is that user email that I provided. That's what ends up creating the workspace and sending it back in. So with all of that, all I need to do is then access the air byte API services. In this instance, we have a JavaScript wrapper that does kind of all that heavy work, heavy lifting to show the widget and then go and actually load the information and then take advantage of all of the logic that we've provided with embedded in this widget and JavaScript code to send that information directly back into Airbite as I showed you before. So now we have our source and data. As an administrator, I can then create processes on top of that to make sure that I measure the the amount of traffic and information going through.

22:04 But I've got myself out of the process of creating and managing all those credentials. I've onboarded all of these customers through that widget or through the APIs. So all of that context can feed to that contextual store that we showed you before. So super quick demo but effectively what it breaks down to is kind of three aspects. The ability to onboard your customers very very quickly. You saw that I created that customer and connected a fake data source but we can expose any source that we provide on the platform.

22:37 And then once you have all of that information as I showed you before it's taken a staging store. In this instance, we had an S3 configuration and allowed customers to create pipelines to that staging store without you being involved from an administrator. The great thing is as an administrator, you can then obviously manage those mappingss and transformations and any of the the data masking that we spoke about before and then finally take that into your workspace.

23:08 Totally. Thanks, Quinton. I think that was like exactly what we wanted to do is just give you all just an insight into how it works, show you how quick it is to leverage and give you a little insight into like what is it that we're cooking up over here at Airb. Obviously, this is kind of just one of the various ways you can leverage Airbite for your AI use case. but the nice thing is that by servicing a lot of AI companies with Airbed, we're actually building out like a fairly strong blueprint as to what we're building next. So, we're going to be doing these webinars monthly where we'll be talking more about what we're seeing in the state of AI, but we're also going to talk about what are the new things that we've shipped to be able to service AI use cases. And for teasers, you're going to hear more things on how we think about unstructured data. You're going to be hearing more things about how we think about authentication for agents, things of that nature. but that said, here are two QR codes. You can always join a conversation we're having in our public Slack channel where we talk more about AI. You can learn more embedded at this link and you can actually meet with our team. I've shared the docs in the channel and I'm also we're also happy to take more time to answer further questions. But for those of you who do need to leave, we did want to just make sure we said thanks for h for joining us this morning or afternoon or evening or depending on where you are. but yeah, we love talking AI. certainly love if you get a moment join join the Slack channel and we love hearing especially what people are building because everybody is building it's like the wild west again now especially the Bay Area it's everyone has this creative idea and it's just good to be part of the conversations totally yeah definitely let us know what you're building we geek out would love to dig in with you and we'll stay we'll go ahead and stay for the next five minutes anyway in case people do have questions questions.

24:54 but otherwise those of you need to log off again. Thanks for joining us. Of course, thanks to Manoj, Don, thank you all for joining us. Thank you Jim for sure. I know it's early for a lot of people, so I'm guessing my coffee after this one for sure. Oh, nice. Get the coffee. You have a big day ahead, so it'll be good.

25:33 All righty. You see people filter out. So, I actually think we're probably go ahead and good to close out. So, Quinton, thank you as always. Anytime to you and I will speak to you all soon and thank you everybody. Thank you everybody. Talk to you soon.