Build Full-Stack Apps on Databricks Lakebase

  • Replit
  • Databricks
Build Full-Stack Apps on Databricks Lakebase

About This Webinar

Every business runs on software, but only a few people can build it. Technical skills, engineering backlogs, and security reviews stand between the people who understand the business challenges and the apps that could solve it.

Replit and Databricks close that gap. Any employee can describe the type of application they need in plain language, and Replit Agent builds a working, enterprise-ready app on Databricks data. Each app deploys into the Databricks environment and automatically inherits any existing security and governance set by the organization.

Now, with native Lakebase support, those apps can go even further. Replit automatically provisions a managed Lakebase database for every app, so every app can both read live Databricks data and write new data back, with no separate database to set up or secure. Automated Preview Deploys and a Supervised AI Migration Flow let teams test safely and approve changes before anything goes live. The result is governed, production-ready apps that go beyond showing data to running real business processes, built by the teams who need them.

Join us to see how Replit and Databricks can help your organization build full-stack apps faster, without adding infrastructure or engineering overhead.

What You'll Learn

  1. How Replit and Databricks connect your enterprise data with the teams who value it most
  2. How full-stack apps open up new use cases across the business
  3. A live demo of building, testing, and deploying a Lakebase-powered app inside Databricks

Speakers

  1. Mackenzie Putici Webinar Moderator Future B2B
  2. Artem Yevtushenko Field Engineer & Data Strategy Replit
  3. Sully Clark Senior Partner Engineer Databricks
Full Transcript Mackenzie Putici Hello, everyone, and welcome to today's special webinar event, build full stack apps on Databricks Lakehouse. Today's session is brought to you by Replit and Databricks and produced by Actual Tech Media. My name is Mackenzie Putici with Actual Tech Media, a future b two b company, and I'm excited to be your moderator for this session today. Now every business runs on software, but only a few people can build it. Technical skills, engineering backlogs, and security reviews stand between the people who understand the business challenges and the apps that could solve it. Replit and Databricks close that gap. Any employee can describe the type of application they need in plain language, and Replit Agent builds a working enterprise ready app on Databricks data. Each app deploys into the Databricks environment and automatically inherits any existing security and governance set by your organization. And with native lake based support, these apps can go even further and supervised AI migration flow lets teams test safely and approve changes before anything goes live. Now today, we're not only gonna talk theory, but we have Replit and Databricks, who will be giving us a live demo of how this powerful team up executes in real time. As you can tell, this can be exciting session for us all. So let's cover a little bit of housekeeping, and then we'll get our speakers up here on… To get the show on the road. Now I've already seen several of you introducing yourselves in the chat tab, but if you're just joining us, just getting settled, if you look to the right hand side of your audience console, you will see a chat interface. And at the top, there are a few different tabs, chat, docs, q and a. So So let's quickly run through those. We've got an aloha from Kauai. So aloha to you, Suzanne. We've got Peru represented, Tampa Bay, Florida, Detroit, Pittsburgh. So if you're just joining us, feel free to give a quick shout out. Let us know where you're tuning in from. Give us an hola. Give us an aloha, hello. And, of course, the chat tab is the perfect place to let us know how you're thinking, how you're responding to the information today. So we love to see all of your comments. Feel free to chime in throughout this session. Hey there, Adam from Atlanta. We've got New York City. Jorge from Texas. Awesome. Okay. So a couple other things to keep in mind with the chat tab. If you have any technical issues during today's session, let us know in there. A quick browser refresh usually is a good fix for any audio, video, slides, not syncing up, not loading. But if you're having persistent issues, reach out there. Myself and Deanna, our webinar coordinator, will be here to help. Beyond that, to enhance your viewing experience, if you wanna hover over your viewport, there are closed captions you can enable. If you're on a mobile device, there's a light mode you can toggle on, which will pair down the console, so great for a smaller screen. And we also have a few things like reactions built in and a catch up button. So if you get interrupted, have to step away for a minute, the catch up button will give you a quick summary of the last ten minutes and get you caught up to speed. With that, we've also got the q and a tab. So any on… Any and all questions for our speakers, throw them in there and we'll do our best to, field those here on the session. And then last but not least, in the docs tab, we've got the Replit Databricks datasheet. So great one pager in there outlining this partnership and all the amazing features and, things that they can do for you there. Last bit of fun, we have a $250 Amazon gift card up for grabs. I'm gonna draw that at the end of the session, So stay tuned for that. You do need to be live and present on this session to qualify for the prize as well as meet the prize terms and conditions, which you can find linked in the docs tab there as well. Alright. And with that, I think we've about done it, folks. So please help me in welcoming to the stage, Artem. Artem Yevtushchenko is the field engineer and data strategy for Replit, and he is joined by Sully Clark, senior partner engineer at Databricks. I'm gonna hand it over to Artem who's gonna set the agenda, and, we'll go from there. So, Artem, take it away. Video Hey, folks. Thanks for joining. Today, we'll be taking a look at building full stack apps on Databricks, like, base, with Replit. And we're we're going to address the problem a lot of us see, today, which is most of the people who want to build software and solutions, on enterprise data are not engineers. And so we'll take a look at how Replit and Databricks helps address that. And we'll take a look at how the Databricks partnership with Replit works and how we bring governed enterprise data together into this, to this Bit Builder persona. So in the next thirty minutes or so, we'll cover this problem, how the integration works, a live demo, security, and we'll jump into some q and a. So a quick introduction for myself. My name is Artem Yevtushenko. I am a field engineer, and I own a lot of data strategy here at Replit. And I'll I'll hand it over to Sully. Everyone, I'm Sully Clark. I'm a senior partner engineer here at Databricks. Before Databricks, I spent about four years working at an analytics company. And before that, I worked, doing farm… Pharmaceutical research, data science specifically. So, my whole career, I've I've worked in and around data, across lots of different roles and use cases. And so I'm incredibly excited to be here today to get to talk about the the better together story with Replit and Databricks. Awesome. Thanks, Ali. And for context, folks, I was also at Databricks in the past, so we all, love the product here. All big fans. Alright. So really quickly, Replit is the development platform where we help both technical and nontechnical users accelerate their work and contribute to product and operations in a governed and contextual environment. So let's wrap it in a nutshell. What about Databricks, Ali? Yeah. So, without going too deep on what Databricks is, the short and sweet version is that Databricks is the data and AI platform. So, the lakehouse is what enterprises have standardized on to govern and run their data. And now with AI, the governed data and context matters that much more. And so, you know, again, I'm really excited to talk about the partnership today, because my career, there's there's always been a gap with with these, the data teams and these end users themselves. And so, you know, great tools have have come and gone in in the last decade or so, but there's always been this this constant thing where, a lot of the the end result gets exported to spreadsheets. And so with AI and with partners like Replit, that gap has been closed dramatically. And so technical, nontechnical users are able to build quickly, in a safe contextualized environment. So, again, really excited to talk about this today, and thinking about where we started, this partnership even a year ago where it was just kind of connecting directly to SQL to the cool places we've gotten to today where now it's one click deploy as a Databricks app or the inclusion of provisioned Lakehouse, which which is what we're going to be talking a lot about today. This is a great partnership where we're we're finding a lot of value for our our joint customers. Yeah. Absolutely. I would say a lot of our successful customers here on the Replit side are also customers of Databricks, and, it's a great better together story all the way. Alright. So, what is the challenge that we're looking to address here? So Replit really came from a place where we saw a lot of teams like sales, finance, operations, HR, needing solutions that really fit their needs. And going through, the limited bandwidth of engineering resources to get those solutions has been historically very challenging. And, an AI, environment needed to be consolidated, needed to be governed, basically a place where folks can go and work, where it's contextual, policies, and all all that good stuff. So with Replit, we bring these capabilities, the ability to build to all of these personas. So you can make your pipeline, forecasting, inventory, headcount, and renewals applications on top of Databricks. We'll talk about in just a minute. And the more we've worked with these organizations, the more we've seen the need for this enterprise, data at really any scale. And we'll talk about scale a bit later in this presentation. And, of course, governance. Right? Making sure that the entire thing is secure end to end. Saul, you wanna talk about maybe any of these use cases and kind of how you've seen this play out? Yeah. Yeah. I think, you know, the… It's the simple principle that, when when you lower the barrier to data, through AI and through what Replit is able to do, the the demand for it goes up very naturally. And so it's, you know, it's funny. A few years ago, you know, people that I've worked with who are very comfortable living in spreadsheets, are now actually building full scale applications, not just dashboards, but, like, applications themselves. So they're able to combine all of these different sources of data, whether it be sales, product analytics, they're they're connecting to MCPs, things that that were were not common knowledge, you know, even a year ago to to these these users, are becoming quite standardized. And they're actually able to build these very bespoke applications that provide a whole lot of value. And, really, again, the the thing that that is is hinging on all of this is that with this partnership, with Databricks, we're able to make sure that this access is governed and the deployment of all of these things is is safe and enterprise ready. So, yeah, it's it's a really beautiful crystallization of what's going on and, again, something that I think both of us are really proud of. Exactly. So we'll take a look at, building on Replit and deploying to Databricks today. That's going to be pretty important for any application or any field where maybe you have sensitive data, or you want to just keep everything in your data estate, among many other reasons here. So this is our… One of our biggest… Bigger announcements. And then to add on to this, not… It's not just you can deploy an application, but you can deploy a full, lake based, instance along with it. So, essentially, what we're offering here is working in Replit, discussing your data, your processes, creating your slideshow presentations, working with agent in a collaborative environment. And then using AppKit, we deploy to Databricks, which means users authenticate as themselves, and the Postgres development, database, which basically is just the data back end for any application you build, is then built up as a lake based instance locally in your data estate. And this gives you basically very easy access and, I would say, native access to a lot of those features, within Databricks. Sully, any Yeah. Interesting, use cases you've seen kind of in in application provisioning in in Databricks today? Yeah. I think, we'll we'll we'll touch on some of the the use cases that I'm I'm most excited about. But maybe as as just a bit of additional background for what Lakebase is, Lakebase Lakebase is is fully managed serverless Postgres that that lives within, Databricks itself. So, historically speaking, you know, the the lakehouse has always been for analytical workloads, whole app processing, which has been incredibly successful, but there's always been a gap where, you know, enterprises have said this is great, but we have this this need for for fast reads and writes in a transactional system, and they've always been very separate. So what Lakebase has done, is really bridge the gap there. And and they're not separate, instances that that live under one hood. They're actually able to communicate between one another. And so whether, it's really fast reads and writes that you need for some transactional point of sales, for example, processing, and and then running analytics on top of that, you're able to do so. Or now in in the world we find ourselves in, having agents that are actually able to run on that OLTP contextual layer is really important. And, you know, some of the things that that are good to keep in mind is that, out of the box, what Lakebase provides is branching. And so it's it's a it's a great experience where you're you're not worried that, you're you're going to destroy the the main database for it. You can have, you know, dev, QA, prod, living in parallel and be able to make changes to those and develop against them without without concern of of making, deleterious changes, full asset transaction, scalability as well. And so it's it's something that, yeah, I will be excited to talk more about in in the use cases themselves as well as in the demo itself. But it's a it's a huge development of our… For us, and we're really excited that our customers are beginning to find a ton of value in it. And and things like this where automated provisioning that that what, Repla does is a huge advantage for us because every every app needs needs a database. And trying to configure an OLAP system to work for that has always been, a bit of a pain. And now with this out of the box, it makes it very easy for these really advanced use cases, to to take flight. Yeah. So… And and something I I I hear a lot of my customers talking about is, like, the security and the governance perspective and the data estate component. Right? So all of that governance that happens in Databricks today, all of that is continued with Lakebase. Right? If I can Absolutely. Yep. Yep. Yep. Unity Catalog is is still, present, for Lakebase itself. And so you're able to, of course, make sure that, with row level security or column level security, whatever it may be, you're able to to deploy these things and make sure that they are secure. No data is leaving Databricks at all. And and it does create this this one cohesive governed experience across all of your data state, whether it be lakehouse or or lake base. Awesome. And throughout this presentation, we're going to take a look at, you know, writing, an application, to Databricks with that lake based instance getting provisioned. But you can always use one of those instances that Sully was talking about about, you know, with with branching and, many versions and testing, and you can always use that as your, default, Postgres DP as well for production type use cases. So the integration is, obviously, quite vast, and we'll take a look at that in a bit more detail, the the flexibility of that integration in the interest of it. Sweet. Yeah. I think this is, kind of speaks for itself really, where we see that when you provision that application, we start with that Databricks, Postgres. And as a managed service, we bring it up for you, and it just lives in Databricks. So I think it should be fairly clear here. So let's talk about use cases. This is why we're all here. So what are teams building? So directly in Replit today, we have an integration, in… With Databricks that spans many capabilities, one of which is Genie. So first of all, when you're having a conversation about your data today, when agent is deciding what to build, how to build it, Replit has its own ability to inspect the data. But what's typically really good for vague questions or things that might need reference, or maybe, like, referral content, we want this knowledge graph that you built up, this this rich context layer you've built up in Databricks to be used to guide all of these answers, and give you these kind of experiences. So we integrate with Genie where our agent uses Genie to get its answers. And then in your application, of course, you can embed these really rich, you know, ask your data experiences, including, genie spaces, meaning you are able to curate really high quality, data question and answer, environments and embed that directly into your Replit applications. And, ultimately, if you want later into that Replit application, publish to Databricks. So think about anything from, delays to bespoke, customer embedded dashboards. It can really be anything. Yep. And switching gears a little bit, is thinking about, forecasting and machine learning powered apps as well. Of course, kind of the precursor to the agentic world we find ourselves in, but but still incredibly important. The reality is is that a a a lot of the world is already using, Databricks or Spark, I guess, for their machine learning workloads. And the really exciting part about this is because this is all already created in Databricks, you're able to reference that very easily, within Replit and create these applications again that are taking advantage of the intelligence that exists within an organization and exposing it to to these nontechnical users, in a much simpler way. And, of course, Lake based adds that to the next level where you're able to actually include something to do with forecasting, where in easy inputs and outputs, you can save that, look at the traces with MLflow and some of these things after the fact, but then also baking in the Databricks functionality that they're really excited about. So it could be, you know, parsing a document. It could be just using the native AI forecast functionality. Or, more recently, we announced something called AI decide, which is a new feature, which basically helps think about what the next best action is. And so, you know, the the TLDR with this is that taking advantage of all of the great work that the data science teams are doing to create these high powered machine learning and forecasting models has never been easier to expose to to nontechnical users, or business users in the past. And, you know, based on on prior life when I was doing this, when I was doing data science work, it was always incredibly challenging to figure out how I shared my insights with my team. And now we can have this one governed comprehensive experience where we're able to develop and then expose this very, very easily. Awesome. And, actually, we'll take a look at, using Lakebase as a feature store in the demo, in the architecture. So that'll be kinda cool to see. Very relevant. So one thing our customers are really excited about is the ability to actually build trust with their customers, and they do this by creating customer portals, where they can get their customers to see their data. And because you can build so rapidly within Replit, folks actually make these bespoke, like, embedded experiences for their customers based on that well governed Databricks data. And they're doing this by building replant apps, do… Doing embedding with JWT, and then deploying this live to their customers. Of course, in the back end, it's all powered by Databricks. And when they're building it, it's secure. They're using essentially OAuth against Databricks to only see the data they're permitted as admins, for instance. And, of course, if they wanted to build something on sensitive data, they can deploy directly to, Databricks as well among other many, many use use cases. So let me, queue up a few other, approaches and a few other opportunities that are opening up to you with this integration. Replit hosts, auto scaling architecture for you, meaning we can we can service a a very large, degree of traffic. Right? It can be small traffic. It can be large traffic. It can be bursting traffic, and we will be able to handle that for you when you deploy with Replit. Now what about those chunkier operations that run your business? Maybe it's processing a thousand documents, batch PDFs. This is where the Databricks integration comes in, and I'll ask I'll I'll I'll have Sully talk talk more about these components. But at a high level, you can have users upload batch documents, track events, and we can trigger batch operations using Databricks jobs. If you want to collect, for instance, clickstream data on your, let's say, ecommerce website and send abandoned cart alerts, we can, of course, integrate that data coming from Replit into Databricks so you can get that timely alert when the user is most likely to buy. On a SQL warehousing side of things, we're querying one of the highest performing engines, out there, and this is where users are getting the best experience. And, of course, we'll take a look even in the demo here where we're calling in, a model. And you have both AI inference, but you also have traditional ML serving, which is, like Sully said, kind of a backbone of, you know, technology today. Anything you wanna add here, Sully? Yeah. I think the the managed streaming and the model serving are are particularly cool and and new and interesting for us. So, the the streaming, has gotten much improved with things like Databricks where, basically, you can take this this, streaming data and have it write directly to, managed UC, within Databricks. And so there's no need to to set up complicated Kafka jobs that that do this back and forth, and you can have basically a one click setup where you're able to run your analytics on top of that. And then, of course, model serving, whether it be foundational models that we have out of the box, but then also, using something like Unity Gateway now, which is the the the governance layer to force guardrails, spend limits on all of this across your users, so you can understand exactly who is is is doing what, and are they deriving value from this, and even create things like rollback in case, you know, you've reached your token limit with, you know, provider a. You can switch to to some… Someone else. So, really, the scale of this, just like Artem said, takes place, in in many different facets within, Databricks, and Replit is able to take advantage of all of the best components. Definitely. Yeah. And and this is why I said earlier that, like, Replit apps are so much more successful, or the the really successful replant apps, I should say, are are successful. The ones that use Databricks are successful. There it is. Because you are able to get this level of of scale, at what is right now relatively easy. The the the actual barrier to entry to do these things that were once very complex and very challenging are now very low, and Databricks makes that completely accessible. So if you wanted to create a… You know, one of the world's most robust applications, in the past, you'd have to be an AWS expert, an Azure expert, and an expert across all these other fields. Now with Replit, you have the expert essentially in your hand, integrating with Databricks, understanding what's possible in there. Of course, Mackenzie is available to you as well. And so you're able to express and describe, projects that would before require a lot more, for for scaling out. Absolutely. So really quickly, this is the Databricks integration nutshell. We've got Replit querying, Lakehouse. So this is going to be actually getting your data in and out of Lakehouse so you can query, SQL warehouses, of course, the lake base, manage Postgres inside your application. The Unity Catalog governance component is really important. We'll respect that with u two m. Genie, integrates into your chats, into your applications, giving you that natural language, component, and Databricks where we can single click deploy applications, into your Databricks data estate. Alright. So when you deploy, as a Replit app, when we say deploy, what does it mean? Well, it's a full stack application hosted on Replit infrastructure connected to Databricks. So this is a publicly external application, a mobile app, a web app, anything from admin and and, you know, vendor access screens, that sort of thing. You can bring your own domain and really bring the full application to the world, based on this data. And then Yep. And and the second option, of course, is deploying as a Databricks app. And so, these are good for use cases where the app is meant to live internally to, the the Databricks itself. And the really cool things about it are that because it runs in Databricks, all of these existing controls apply automatically. So network controls, egress, and allow lists, it inherits UC permissions out of the box. Users will sign in with their Databricks identity, so the app is only available to those, who have been given access within Databricks. No data is leaving Databricks, of course. But then the the really big important take home message with this is that you can have all of the functionality that Databricks provides. So there's no there's no sacrifice you're making here in terms of the the richness of the features that are available. We've we've talked about Lakebase, so, you know, transactional reads and writes for this. You can bring in Genie for conversational analytics, data smart coworker, model serving, foundational model APIs, and and gateway to govern all of this. So, really, both of these paths, are are different for… From a use case perspective, who who is the the end user of this and who… Who's it meant to be exposed to. But the the story is the same, that you have a really, really solid world class foundation that you're able to build on top of and satisfy these different use cases. Exactly. Alright, folks. So next, we'll transition into a quick demo, and let's talk a little bit about the use case. So a quick show of hands. Who stood at a baggage carousel after it stopped and your bag never came? So in Japan, plenty of travelers skip the carousel altogether. They send their luggage to the hotel a day early and take the train without nothing… With… Without, anything in their hands. Nothing to carry. It's great as long as the bags actually beat you there. So in our example company, Promise Desk, they run this kind of service. And tomorrow, a tour group is headed to Tokyo from Sapporo, and their 200 bags are scheduled to be picked up. So it's Monday morning at 9AM, and this lands on your coordinator's desk. The hotel and support just sent over their 200 bags for tomorrow. They need everything by 10:30. Are we gonna make it? So let's find out. What we're going to do is use the data that's already in Databricks, bookings, delivery plans, forecasts, and a machine learning delay model in Databricks. And the coordinator just the coordinator just can't get at it without a little help. They're not the most technical user. So luckily, they just got Replit, and they're using it for internal business operations, various slideshows, many other things. But, of course, it's connected to their data state, and they're going to ask certain questions like this. So we're going to go from a one off prompt to a scalable solution for your customers and internal teams. This is how a lot of things in Replit start. The user will ask a question, and they'll find that there's a better way. And so what we'll take a look at is how travelers, can check when their bag will arrive. Basically, before they book, they can check using that machine learning model. That side will be deployed to Databricks. I mean, that side will be deployed to Replit. And then the operations team that manages all these bags, that will be owned in a Databricks app where they compare plans, approve changes. This is your operational portal. So let's start where the coordinator did by asking a question. So we'll open up Replit, and we'll see this chat box. So right away, we'll ask the question. A hotel in Sapporo just booked us for a tour group's 200 bags. We picked them up in Tokyo this morning, and and they need to be at the hotel by 10:30 tomorrow. Are we gonna make it? So like we said, the data is already in Databricks. So all we have to do is just essentially orchestrate, some really good thinking. We have to look at the data, analyze some metrics, understand the problem, and Replit is able to go into that data, introspect on it, analyze it, and, of course, integrate with Genie, for more in-depth, maybe, contextual awareness. So in our case, agent will think for a moment, run some queries against the SQL warehouse, perhaps use the AI model. We have a machine learning model there as well. And it'll tell us that, actually, there's snow coming, and your ferry route that you had planned on the books is actually at risk. So what you have to do is move the… Move those bags to an extra air run-in order for that not to be late. So what should we tell the hotel? The bags are on track for 10:30, and we're not gonna tell them about dispatch moving that. We'll take care of that. We'll own that process. Right? And agent also asks a nice follow-up question. This is how a lot of things get started in Replit. A user will go into Replit, ask a question, have a conversation, use agent as a design partner or a thinking partner, and Replit will suggest, you know what? This process seems maybe broken, or this process seems like it could be optimized. And so Replit will suggest a couple of options for us given that it knows the context of what it can do. Why don't we start with a customer site? Osire and Replit. Travelers sending a single bag enter the route and deadline. They get an arrival estimate. One, they can book. We know we can do this because we have the data. We have the model already in Databricks. And two, we can make a dispatch tool directly in Databricks. So your teams can compare plans, see the weather along both routes, approve extra runs like this one here today. And Repo will ask you a question. Should we build the customer site now? Yeah. Let's go ahead and do that. And so here, you can see Replit suggesting a pretty comprehensive plan for starting a new project. And with a little bit of demo magic, we're gonna hop over into another chat where our upload has already gone and done this. And so we can click on the project that it's created right there, and we can find the application, that it was building. So Replic creates this app for us. It understands the problem, and so the agent, prompts this application to start and, of course, is able to integrate with it. And what we can see in this application, if we hide the sidebar here, that it already integrates deeply with Databricks and that the model APIs are integrated, SQL, and all of that. But you didn't have to set that up. You didn't have to prompt that. It knows contextually where the data is and how to access it. So I can, for instance, choose, a couple of words in this area, suggest when I'm leaving, like we talked about, for instance, checking, if a Tokyo and Sapporo route will be on time, and then we can see that risk that we saw earlier. Right? It's coming from the same data source. It's coming from the same machine learning model. And we can choose, for instance, the air route. It'll cost a little bit more. Or we can choose the slower route that's at risk and just say, sure. I understand that risk. And now Replit created a portal where instead of having to reach out to me, the operations manager, I can just go into the application. Cool. So the user has uploaded, their information or has sent the information for scoring. But what about a hotel or a large group where you might be managing, you know, 200, a thousand, or more bags? That's where the Databricks jobs integration with serverless comes into play because it does give us that scale for building a true enterprise application. So for instance, let's say I'm a hotel or a tour group, and I want to analyze that all of my bags, all of my customers are going to get there on time. So I can plan this delivery, and what the application will do is make sure the schema of the document is correct. It's going to upload that document to a Databricks volume, kick off a job, and, of course, start preparing the data, scoring the data, and placing every bag on the best route. Okay? So here we can say now it's preparing. And let's take a quick look at what it looks like in Databricks, and you'll find that it looks very similar. If we go into Databricks jobs on the right side here, we can see here is a run, and you can have as many of these… Almost as many of these as you want concurrently. And we can see that this lines up basically one to one to what we have in the application. Awesome. So the customer will come back in a couple minutes, and the data will be all processed, and the report will be waiting for them. So you can see here now it's scoring, and it's ready. So let's take a look at this one here that's already been run. So we're waiting on the dispatch to confirm some of these bags because, as it happens, a few of these can't make it for the same reason we saw earlier. Right? Most of these will make it, but there's that, snow conditions coming our way, and it just won't make it. We have some eight bags from Tokyo, 27… Tokyo, eight, six bags of Osaka, four bags from Osaka. Whatever point is, it's not gonna make it. So, this use case scales out to a lot of other things, though. Think about, for instance, uploading maybe loans for for fraud evaluation, or it could be, a bunch of PDFs with invoices, legal documents, and you want to parse and extract that information, load it into the system, score it for maybe approval or, you know, flagging something for potential fraud like we talked about. This kind of system can scale out really neatly with Databricks because you can create these interfaces, these human in the loop interfaces, on really scalable solutions. And so we talked about creating that dispatch application for, Replit. But before we move in there, let's talk a little bit about publishing. You've created this app, but so what? So in Replit, we can go into publish, click on who can access your app, and say public. If I go to settings, I can even add my domain. This is my company, and I'm making this, application for my customers. I simply say I want this to be public. I'll go ahead and run security scans. I'll add, active monitoring. We can enable analytics and deep analytics to basically track user progress throughout the application, creating a feedback widget so users can tell us if something's not working well for them, and create a really great user experience on the front end. So we… When we click publish, it'll go ahead and make that available, well, to the Internet. Now while it's running, let's take a look at another application we built in a very similar process. This is a front end for… Or, obviously, the the back end for PromiseDesk. Try to move this guy over. Here we go. And this is kind of a triaging solution where we can, move a 120 bags that we saw earlier from the ferry so that the risk goes from a 120 to zero. Right? We had the customer already upload that information before, and now we wanna triage this. And so here we have information. We have, for instance, our, ferry routes. We have our snow disruption, and then we have a log of who did what. Right? So we need to track this against Lakebase. In this case, we're doing this locally with Postgres, but we're going to deploy this to Databricks. Well, how do we do that? Same button. I've already published this. In your case, it would say publish. In my case, it would say republish. Let's take a look at it. When you publish your application, it'll go through the same steps, security scans, making sure everything looks good, packages it up as a Databricks application, gets lake based provisioned, everything goes. And then you simply click on the link. And let's take a look at that. Let's take a look at that. That is your Databricks application running your, app that you just designed and deployed. That's as easy as it gets. Now I can go ahead and say in here. Okay. Sounds good. Approve the fix. Yep. Look at that. It knows exactly who I am because we're using on behalf of authentication in the Databricks app. It authenticated me, and all of those records in Databricks are being provisioned here. So let's go back to our application. And one thing I saw… Oh, this is great. It actually found a security vulnerability that it needs to fix. So it won't fix it without those security checks. Let's actually go ahead and open up agent. While we're working, we'll have agent fix that for us. So the other thing I wanted to point out is how this works behind the scenes. Right? What did I just show you? Well, Replit Agent builds both experiences. It builds the customer application, hosted on Replit, built on modern React APIs. Let's actually go ahead and hide this, for now. Where we have a working preview verified, and then we can simply publish to the Internet. We also have the internal dispatch app, a separate project. It's a hosting target. It's, built to Databricks applications plus Lakebase. So you have a great OLTP solution for a very important application. All of this is built against Databricks on operational data integrating with the machine learning shared delay model, meaning this single model in Databricks is used across all of these solutions, and we have the scalable planning option as well using scalable jobs. Yep. When you use heard him. I just wanna, color that that previous slide in a little bit more as well where, previously, we we talked about how these two different deployment methods, whether it be publicly external facing using replica to do the hosting or as a Databricks app, The the demo that Mackenzie just walked through combined both of those into one experience, which is a really complicated thing to do, and he made it look as as easy as possible. So, the ability to to connect to this one single source of truth and satisfy these external users who are able to, like like, showed in the demo, route these bags successfully taking all, advantage of all of the intelligence that exists within Databricks. But then also for those dispatchers who are working internally, who definitely don't need this to be a a publicly facing, experience to be able to have that same source of intelligence but actually do the the communication and dispatching through there. So just wanted to highlight that. This is a very challenging thing to do in practice that Mackenzie made look incredibly easy. Absolutely. Well, Databricks made it incredibly easy. Right? That's the beauty of Replit. You can do these things in in evening. Right? So what did Replit build for me? Well, I described what I wanted. And so what Replit build behind the scenes is actually most of the Databricks stack that we saw here today. I I did wanna call that out. Obviously, we're not using real data. Right? We're not using actual, you know, baggage data from Japan. The model that we're using, that had to be, custom created as well. Well, as you're looking at this architecture and these individual components, I didn't wanna call out this was all Replit. I described my solution. I described that I wanted a model to be served, that I wanted it to be registered, And it was able to build this essentially all by itself and, of course, you know, using that Genie integration as well-to-do some work. And in that collaboration, built a functionally working MVP in an evening. So as a Databricks admin or as a Databricks power user, this can also be an entry point into a lot of the work that you're doing as well. You've got Genie code when you're in the platform, but you don't have to lose that capability when you're in Replit. So what did it make for us? Well, it created the customer inputs that we saw earlier. Right? The user interface. It created the back end. It it built the features and called the model. Right? It took those inputs the user made, structured it so that it could hit actually… I said I say SQL warehouse here, but Lakebase actually is, what we're using behind the scenes, doing a feature lookup, hitting that model, with the new weather data. Right? We talked about how we're using live weather data. So we do a live feature lookup, get the weather, join that on using this app production alias model, sending that back and telling the user in just a second if their bag is on time. On the other side of things, we took a look at integrating with volumes, taking CSV, PDFs, whatever, running AI functions on them if you want for, extraction or just parsing them out or simply feeding them into a pipeline. We run that volume essentially as an ephemeral location for landing that data, preparing that data, using, Spark to score that bag. Data in batch, and then allocating capacity and persistence. This is really complex to do, but Databricks makes it very easy. That data gets committed whether to Delta and Union Catalog, Lakebase, as a target only. Doesn't really matter. Then we read that data. We pull it and show the customer. So what did we just say… What did we just see? We saw somebody ask a question. In Replit, it suggested there's a better way. We can build these interfaces. We can build these better operational methods for travelers to check if their bags are gonna make it on time, for hotels and tour clients at scale, answer the question for their entire fleet of, customers. And then on the backside, we made an operational dashboard, an operational form or platform built into the data platform that you already use and trust today. And this is really the beauty of this integration. So the idea is simple. Build and repl it, zero trust isolated sandbox with really robust, security scanning. Lot of great controls and governance of what the agent can do, what it cannot do, and that integration with, Databricks. The apps are shipped into your tenant. The data never leaves it. It's secure. In Databricks applications, unit catalog governance permissions are enforced at runtime. You can use service principles, so everyone shares the data or have it based on their user, well, on behalf of authentication. And, of course, it's SOC two compliant and continuous… Continuously monitored. Now I think this is actually talking a little bit about Replit. This does make it sound like we're actually doing these things on Databricks. We actually do a lot of these things in Replit also. Sully, do you care to talk a little bit about maybe the the authentication stuff? I mean, I know it's boring, but, like, knowing who you're… Who you are and what data you should get is pretty important. Right? And then, of course… I mean, user certified in all sorts of ways. Right? Yeah. Absolutely. I think, you know, there's there's different ways to to satisfy each of these use cases. And so thinking about deployment as Databricks apps in particular, it's really important to to think about what that would look like from the end user perspective. And so, the the best example Artem Artem showed it very clearly is on on behalf of authentication. So that basically says, who is this user? Are they in the Databricks organization as the first check? Great. If so, do they have access to, one, use this application, use these resources, and, use the the data that's available within the application as well. So that's that's, I think, a really important starting place for all of this because, governance, of course, is is a huge priority for for both of us, and making sure that the app doesn't doesn't accidentally get exposed to to internal users who shouldn't have access to it is important. Similarly, though, authentication with m to m, so machine to machine using the app service principle as well, is also a very good starting place where, if there are shared resources, that that are, you know, needing to be used throughout the organization, that is also a very viable path just to secure, just not taking advantage of the OBO for that. Perfect. Alright. So, we'll jump into q and a in just a second. But, you know, couple of links here for your convenience. So how to contact us, repla.com/partners/databricks. Request access, get your setup guide. We'll work with you, to get you started. And step two, go to developers.databricks.com. You can go ahead and take a look at your prebuilt prompts, and a bunch of other resources to get you going really quickly. So it'll take a second here. You can go ahead and screenshot this page if you like, and let's jump into q and a. Alright. So I see, one question. What does Replit automate when provisioning Lakebase, and what remains our team's, responsibility? Yep. Great question. So when you deploy an application to Databricks with, an application that uses Postgres behind the scenes, we will automatically provision that lake based instance for you. We will hydrate that data if it needs to be hydrated, or just make the schema, and you can just start using it. From a management perspective, Mackenzie, I mean, is there anything else the customer has to do? I think that that's largely it. I mean, the the setup for it is, you know, when you set up the the connection initially, from Replit to Databricks, but, still finer grain controls. So, you know, what is the access that that users have on top of that, setting up, the the auto scaling or any other capabilities that you're looking for in there, kinda tweaking the the functionality with it. But the important thing is that Replit does the automation, of the the creation itself and connecting to the app with short lived credentials, and and no stored passwords. So after that, it's it's totally up to the user if they want to create branches from from main, whether it be, again, through dev or QA. They're more than able to. You can do that in Replit or or in Databricks itself. And then I guess the last thing I'd say too is that if you're looking to do, say, a lakehouse sync, which is taking that data that exists in Lakebase and writing it back to the lakehouse itself, you could set that up. You would have to set that up in Databricks, but it is a very simply a one click experience, where you basically choose the destination you want that to land in the lakehouse itself. Nice. Yeah. And if they want to, I mean, folks can, write data directly into, like, base from production. Right? It's just, like, a standard endpoint that they can just connect to. Right? Exactly. Yeah. Perfect. Yeah. So if you got a production application, feel free to do that. And, also, federation is another cool one I've seen where users actually will connect to their Postgres instance of, like, production application and just read that and, you know, mesh that up with some other data they have in Unity Catalog. So that's been another Yep. Interesting design pattern. Absolutely. Alright. Let's see what else we have here. How are permissions handled? Does the app use each user's existing Databricks access or its own identity? Mackenzie, do you wanna talk a little bit about Databricks apps quickly here, and then I'll Yeah. Yeah. Yeah. This this was, kind of featured as as the the… One of the last slides there. But, U2M or m to m are are the main ways of doing it. So, again, user to machine, which is OBO, or m to m, which is using the app service principle. Each app has its own service principle that's created by default with this. That doesn't mean you have to use it, but that means that you are granting permission to that application for the resources, for the data, for whatever it needs in order to function. But then with UTM using on behalf of, it is the more, I don't wanna say more secure option because they're both secure, but the more fine grained option, where if if you're looking to, again, restrict access to certain things, simple example is, you know, if it's a a a franchise performance app and you don't want the East Coast teams to see the West Coast or vice versa, using OBO or or UTM for that to to make sure that if you have row level security in place for your data itself, there's no concerns for that. And then the other cool thing with that is if there are, of course, you know, national managers for that, they're able to see both of those because their Unity catalog permissions would would permit them to do so. So, again, the TLDRs that replit handles all of that very well. With with some careful prompting, I'm sure you can figure out the the exact use case, for for what you're looking for and which authentication pattern makes the most sense. Yep. Absolutely. Yeah. And and in Replit, it works the same exact way. We have two integrations. If you search Databricks in the integrations bar, m two m and UTM, exactly like we just talked about. Set up an UTM, you'll need your Databricks admin to do what's called an OAuth callback, but just make sure that you have that Databricks admin, available to help you set up set up UTM. And for m two m, all you need is a workspace admin, and then they'll be able to help you set that up as well. Similar situation, m two m gives you, kind of blanket governance. Everyone kind of shares that same permission in the workspace, maybe like a shared schema, or or shared, like, tables, typically, like, read only. And, you two m is going to be folks using, logging in, getting access to resources that they can get access to. Right? So Yeah. Fine fine or grain to control. Alright. I see a third question here, and it's, can business users build these apps themselves, or will they still need help from engineering and data teams? So I can lead this really quickly here. So business users can typically get very far, and we can help them get even farther, by using something called, like, workspace level skills, for instance. If you have any best practices, you can, of course, always incorporate those best practices into, like, a skill. And so everybody in the workspace will incorporate, will will be able to leverage that. So if your data team wants to partner, for instance, with your business unit and say, hey. We wanna build these apps. We wanna make sure that it's scalable, it's it's secure, and that people aren't just doing, like, shadow IT type stuff. Right? You wanna make sure that you're… Maybe you chat with your data team, you chat with your engineering team, what have you, maybe your Databricks admin, and, have them describe essentially, you know, what what catalogs, what schemas should you use, what are resources that are available. And then through, skills or another integration, maybe, like, they have this documented in Notion or Confluence or anywhere else. Right? You can just simply point agent, to those resources. Those integrations are, native out of the box. And then that bridge becomes a lot easier for business users so they don't have to engage, like, engineering teams and data teams all the time. Now I will say, you can bring these demos to data engineering and demo, data engineering and data teams and say, hey. Here's the idea. We've done prototypes. We've done recorded sessions. We've demonstrated this is what we want. Help us build it. And, typically, with that handoff with the engineering and analytics is very smooth. It comes down to making a commit and repl it, and now they have all their stuff in GitHub. Any any thoughts on that, Sully? Yeah. I think, sort of from first principles here, it's it's, very hard to trust the the quality of an application if the data that is powering it is in a shaky place. Right? And so, I I I always recommend to to build a strong data foundation. So Databricks, preaches the medallion architecture, making sure everything is gold layer, in the end, ready for analytics data on top of that. If you can build out semantic clinicians for these things as well, that that makes it all the more all the more better. But there are things, from from the Databricks perspective that can make these things easier, like spark declared pipelines, Genie… GenieCode pop up all over the place in Databricks now to help expedite that. But, to Artem's point about trying to limit shadow IT, it is definitely advised to go through the the data teams and engineering to actually codify and make sure that the the data foundations are are good for this. But, at the same time, there are there are probably… If you're getting started with with Replit and Databricks together, you can create what I would call a a self contained app where there's… You're not reading in any data necessarily, but, sure, you're writing to it. So maybe it's a a task list of things or, you know, the the the very cliche to do app, just to make sure that things work, then that is, you know, totally something that you can do without the need for for a a data team to get you started. But for these apps that that we're preaching for for what Artem showed, definitely makes sense to have well modeled data with with semantic definitions in place. Definitely. And I think that last piece you mentioned about the semantic definitions, having an agent know where to look, you know, working with Genie. That's so important. Like, you you need… Folks need… You know, we wanna make sure that, it's, you know, commented and that that stuff. Alright. Do we have time for last question, or we have time here? Well, let's let's do the last one. Why not? Alright. Perfect. So we already have dashboards in Databricks. What can we build with Replit in Lakebase that goes beyond reporting? That is such a great question, and it… It's a really fun one. I I hope what we saw today, in the demo where the user was able to take action, on that information, we hope that that's a pattern that you can use, globally, whether it's with your customers or internal operations teams. We want you to take action on information, and a dashboard is is is great. We love a dashboard. Right? But being able to go beyond it, maybe integrating with Slack where you live today. Right? Maybe we have, for instance, the Slack integration with Replit. I think Databricks also has one. Being able to create mobile applications that you can bring with you into the field, so that you can take an inventory to scan things, take an inventory, do data entry, take a picture, annotate, whatever it is. Right? Repl dot it doesn't just do web apps. It creates any kind of artifact. Right? So this can be something that runs on prem on a COBOL machine. You know? I'm joking. I mean, of course, you can't do that, but you could. So let's go beyond data and go into into making the data decision making. Sully, any any Yeah. Like, thoughts and closing thoughts on that? Yeah. I think I… I've always considered, you know, having worked in the analytics space for for a long time now that, especially with with where we find ourselves currently, dashboards feel a lot like reading a newspaper where you're getting some information, but then what you do with that information, is is unclear. And, hopefully, what we've shown today is that you can take it to the next level. And and, I mean, Artem and I, both speaking from experience, we're building a lot of apps and and not really dashboards. Sure. They have rich analytics that that go along with, the the functionality, but the functionality is is the main thing. I I want to get some insight from the the telemetry that I have or or things that I need to do with with, customers. But then I need to take action on it. I need to send that email. I need to to bring in the the transcript from a call that I had. And that's where the… This really rich, experience is brought together. So it's not, dashboards aren't good. Dashboards are obviously, very helpful. They're not going anywhere anytime soon. But, we… I think both of us want to to challenge everybody to think about, some of the things that they've they've learned today, and how they can apply that to their their day to day and how it goes from just, again, being I'm reading some information, but now I'm actually able to do something with that. So, but, also, if you have dashboards already, I would highly encourage you to say to Replit, hey. I have this dashboard that does this. Can you help me make an application out of this and give me some suggestions for where I can improve this, what what I can do on top of it? So, it's great that you have dashboards, and and let's let's get building on top of it. Exactly. Perfect. Thanks so much. Mackenzie Putici Alrighty. Artem, Sully, thanks so much. I know we are at the top of the hour, so I'm gonna let you guys jump. And everyone out there in the audience, thank you for sticking with us. I know it's been a great conversation. The demo was fantastic. And, for those of you who asked some questions and we didn't get to them today, don't worry. We're gonna hand these over to, Replit and Databricks and make sure they get any and all information that you're looking for. As we are about to close out here, we have a few things to cover, and then we'll sign off. So quick reminder to hop into the docs tab there if you wanna get that, Replit Databricks datasheet. If you guys wouldn't mind, we are gonna launch a survey, from us here at Actual Tech. Just getting a quick bit of hot feedback about this webinar, what you liked, what you didn't, maybe spot some areas where we could improve. And then we are also gonna give away our prize. So our lucky winner today without further ado is Jerome Kiefer from Colorado. So, Jerome, we're gonna reach out, by email right after this, let you know how you can claim your prize, and then we'll get it over to you right away. With that, big round of thanks to everybody. So, obviously, thanks again to Replit and Databricks for making today possible. Thank you to our speakers for being here, sharing everything with us, showing us under the hood in that demo. And big thanks to all of you in the audience for taking time out of your busy schedules, being here, and, hanging on till the end with us. It's been really great. Thanks for your questions. Thanks for your comments, for all your feedback. If you do wanna check out any other webinars, events.actualtechmedia.com is the place where you can check out that QR code up on your screen. And until our next webinar here again together, I'm wishing you all a fantastic rest of your days. Take care, everybody.

Please provide citations & attributions for any information you generate from this content, thank you.