Sponsored by: Sigma | How Sigma Makes Databricks AI Safe and Scalable for Every Business User

Summary

Sigma brings the power of the Databricks Data and AI platform to everyday business users through a familiar spreadsheet interface layered with natural-language AI querying, allowing users who have historically relied on analysts to interact with live Databricks data using plain English.
Dotmatics embedded Sigma into its Luma lab orchestration platform to give scientists a governed interface for managing lab workflows and analyzing scientific data, replacing spreadsheet-based analytics with a scalable solution backed by Databricks.
Sigma's roadmap includes MCP server integration for contextual AI questions over live data, automated anomaly detection with scheduled reporting, dynamic conversations over live Databricks data, and agent-to-agent connectivity via APIs and MCP.

Chapters

FAQs

What is Sigma and how does it work with Databricks?

Sigma is a business intelligence and analytics platform that presents data from the Databricks Data and AI platform through a familiar spreadsheet interface, enabling business users to interact with live warehouse data without writing SQL or Python. This video describes how Sigma's natural-language AI layer allows users to ask questions in plain English over live Databricks data, bringing AI-powered analytics to daily workflows.

How does Dotmatics use Sigma and Databricks together?

Dotmatics built its Luma lab orchestration platform on top of Sigma and the Databricks Data and AI platform to give scientists a governed interface for managing lab workflows and analyzing scientific data. This combination replaced spreadsheet-based analytics with a scalable, governed solution that embeds AI capabilities directly into the scientific data pipeline.

What is Sigma's MCP server integration for Databricks?

Sigma's roadmap, as described in this video, includes an MCP server integration that allows AI assistants to ask contextual questions over live Databricks data through Sigma. This would enable AI agents to retrieve governed analytics context from Sigma's layer without requiring direct access to the underlying warehouse tables.

How does Sigma support embedded analytics for business applications?

Sigma can be embedded directly into customer-facing products, allowing companies to offer governed AI analytics as a native feature of their own software without building a separate BI layer. The Dotmatics Luma example in this video demonstrates how embedding Sigma into a product delivered AI-powered data exploration to scientific users as an integrated part of their application workflow.

Full transcript

[00:07] Hey everyone, my name is Alex Shueri. I am a director of product at Sigma. >> And hey everybody, Michael Fritz. I'm a principal on the product team at Dmatics. >> Yeah, today we're going to talk about how Sigma makes data bricks AI safe and scalable for every business user. So we're going to go into the details of
[00:19] it. Um there's a quick overview of the agenda. I'm going to talk one slide just data bricks AI and the opportunity that sigma presents uh as as partners and then uh Michael is going to talk about
[00:32] use case with sigma and data bricks. We're going to talk about AI um and what we are working on uh from a sigma perspective. I'll show you a quick demo um and then we'll we'll open up towards the end for Q&A.
[00:45] So with that let's talk about data bicks AI and the opportunity this this partnership kind of presents. uh data bricks has made a lot of strides and and innovated in uh kind of this new era of AI. uh they have worked on Genie spaces
[00:58] like warehouse-based models uh supervisor agents you might be all familiar with um they have UC metrics they have trusted enterprise data that that people have loaded as part of data bricks experience um sigma
[01:11] essentially brings that power of warehouse to every business users right we've kind of go beyond analysts beyond um the data scientists of the world into like daily workflows people uh who've uh
[01:23] who've historically struggled with writing SQL Python on and so on. Sigma presents a very familiar interface which is a spreadsheet to them and now with AI you can just have English as the this the mode of operation. Um and so this
[01:37] combination of these two products have been super powerful and a lot of our customers have got a lot of value not just for internal operations but embedding Sigma into their products which Michael is also going to talk about briefly um and and how they have
[01:50] generated a lot of value for their customers as well. Uh so with that I will pass uh over to Michael to talk about Dmatics Luma. >> All right. Thanks a lot. So uh yep everybody my name is Michael Fritz uh and at Luma I'm responsible for luma lab
[02:04] connect which is our uh lab data orchestration platform and instrument integration tool. But first right a little bit about who are we in short we are a company by scientists for scientists. Uh over the years the
[02:16] company built uh and developed some of the worldleading tools for processing scientific data and uh statistical analysis like graphpad prism omi uh genius biologics suite etc. But one of
[02:28] the things that we recognized and one of the challenges that I'll talk a little bit more about was that all these tools were great in their very specific domain but didn't have an overarching way to connect and integrate across the systems which is why we built luma which I will
[02:40] spend a little bit more time talking about. Um, also was acquired by Seammens about a year ago now. So, while we had historically focused much more on the research space, um, specifically in life science and CNM opportunities, Seammens
[02:53] is now helping pull us further downstream into development, manufacturing, text transfer, and scaleup locations, uh, as well as pulling into new industries and verticals, really delivering the first industry agnostic tool for taking
[03:05] scientific research data and progressing it all the way from the initial molecule and research stages into development. Um so what is that challenge that we faced? Well, anybody that's worked in a research-driven organization understands
[03:17] how disconnected labs can be. Uh people that are working on biologics research are highly separate from people who are working on chemistry research. Even within an organization that spans both therapeutic areas are often separated by
[03:29] organizational boundaries. There's tons of externalization out to CRO which creates their own integration challenges. And effectively what you are left with is just a massive massively disconnected uh data system with silos
[03:43] everywhere. And if anybody's at this conference, you probably appreciate that that is not the type of data foundation that you need to be able to enable an AIdriven future. Um specifically at life science companies, we're seeing a lot of
[03:55] shadow IT organizations and shadow IT that's being built to develop and harmonize and automate uh different workflows which leads to inconsistencies across groups. uh inconsistencies and challenges of dealing with vendor
[04:07] formats etc and ultimately leading to lower quality data platforms which in effect delays analysis um it just causes reporting headaches. So our goal and our thesis with LUMA was
[04:19] to say how can we take the world's leading lakehouse technology data bricks and bring that to scientists because again you know while this or group and this room might have expertise in data
[04:31] engineering and computer science and be able to take the technology that data bicks has as we have and build a platform and product off of it. An average bench scientist is not going to have that. So we took data bricks and we packaged it up with a lot of other
[04:44] industryleading technologies. um Apache Knifi, Kafka, you can kind of see a a a subset of the technologies that we use at the bottom of the screen and built this platform, but we were still left with that final model problem where
[04:57] organizations could conceptually understand this, conceptually understand the con the idea of a contextualized and harmonized data set, but it didn't really feel real until they saw their data on a dashboard and visualization
[05:09] tool. So, we were trying to figure out how can we do this scalably? we've built this amazing SDLC like app configuration, deployment model and governance model. How can we do the same thing for experiences? And we were fortunate enough to meet Sigma uh
[05:22] actually at DIS three years ago today and kind of realized it was that missing piece for us. Um their approach to building workbooks and allowing us to integrate that into our config model for portability was perfect and perhaps the
[05:35] biggest differentiator for us at least was the plug-in framework that they also have. So while most BI platforms could specialize in things like you know bar graphs, line graphs, potentially some geography mapping, what Sigma really
[05:48] allowed us to do was take those years and decades of experience building scientific specialized tools like sequence viewers, chromatogram viewers, contour plots, and package that in a way that allowed us to deploy that in the
[06:00] visualization layer, effectively allowing us to build a scientifically informed visualization layer that worked with our data modeling capabilities. So we got to focus on the parts of the process that we knew best and Sigma helped massively accelerate us um in
[06:13] that final mile. >> Yeah. I just saw the demo of their product. They've used a lot of plugins. >> Yes. >> As part of their experience. >> Yeah. In addition to the visualizations, the plugins have also allowed us to really deeply integrate at the data
[06:25] level with our ecosystem. So it's been >> um a true game changer for us. And no other BI tool on the market had anywhere near that extensibility for us, which is how we were able to really adapt it to our use case. That's pretty solid. Yeah.
[06:39] >> So at the high level, right, how Luma works, um, Lumal Lab Connect integrates across the disparate data sources and systems that a client might have, any instrument can flow through either our filebased layer or our API based layer,
[06:52] integrate with a variety of processing tools, integrate with other systems either that we own or third party tools like ELN's, limbs, etc. connect with CRO's and basically create this massive harmonized contextualized data uh store
[07:06] that actually creates context across those different pieces of data. So if you saw Ali's keynote yesterday, no longer are the AI tools going on a random walk trying to explore this. Luma provides a structured modeling layer
[07:19] that effectively creates a semantic layer specialized in scientific research data that organizes this data and then we use Sigma to present that in ways that make sense to users and allow them to not just see the data from a BI
[07:32] perspective but really truly interact with it, explore it and make decisions and commit those to the experimental record. >> Yeah. And just to kind of elaborate a little bit on that, like these users that you're exposing these dashboards to
[07:44] are not like super technical users. These are scientists as you said and you're basically meeting them where they are and what they want as part of your product. >> Yeah, absolutely. It's it's not the computer engineers who could build these pipelines themselves. It is it's a
[07:57] scientist and they care about the science. They they might have some technical expertise but the ability to democratize these capabilities has been again part of our underlying thesis. It's something that Sigma helps us deliver. >> Very cool.
[08:09] >> So on the right side here you can actually see a little vignette of like one of the workflows that we have. So this is specifically a retroynthetic chemistry workflow. A scientist submits a molecule to uh to domatics. Then we
[08:22] integrate actually with Millipor Sigma's Cynthia product to basically find out what are the 10 20 50 different sets of reactions that I could take to develop this molecule. One bit of AI enablement to help scientists. Luma then takes
[08:35] those potential reaction pathways, enhances it further with AI based off the experimental record and IP of that company, ultimately creating recommendations and allowing companies to decide which of these reactions are we actually going to try and create. Um,
[08:49] some of this is driven by a few different plugins as I mentioned earlier in terms of the actual integrations that we've set up and then also you can see our lineage viewer plugin again something that you don't really need if you're not a scientist but is is really fundamental to what we have built. Um so
[09:03] ultimately what the platform does is help create that centralized store of data power this interactive analysis and sigma uh give us abilities for instance in here as well to add AI based summaries of experiments and workflows
[09:15] and really ultimately bring these capabilities to scientists as we've talked about and also because it's all uh configuration based we've built in a governance foundation that really allows organizations to scale their AI
[09:27] initiatives in a turnkey way um which is really a major blocker that we're seeing in industry if people try and go if they try to go direct uh to an AI future without really thinking about governance
[09:39] upfront >> truly. So ultimately what does this mean? Well, we've helped organizations go from heavily fragmented organizations to providing that strong data foundation
[09:51] that they need to enable their AI future. By creating automated reusable pipelines that fit a variety of different use cases, we've helped reduce the impact or need for shadow IT organizations and ultimately accelerated
[10:04] uh collaboration with CRO's improving data consistency and trust and again ultimately helping deliver the scalable foundation as well as some uh AI based tools ourselves allowing companies to
[10:16] continue to accelerate their research workflows. But that's kind of what we've done today. This is a little bit forward facing now. I don't think we're stopping here. I think we're actually really excited for some of the work that Sigma also has coming out. So, as you saw from
[10:29] that experience, right, the the types of experiences that we are configuring in Luma >> are highly highly complex. So, we are very very interested and uh eager to get our hands on some of the AI enabled workbook building that the team has been
[10:41] working on. >> Um just to continue to help accelerate our go lives and also to start experimenting more with conversational AI in those experiences. Um, for those who aren't scientists, a typical high
[10:53] throughput plate might have like 384 wells in it. And in a high throughput setting, you could easily be running 200 plates per day per instrument. That's over 75,000 data points in individual experiments that need analysis. Now, if
[11:05] you are a bench scientist, you probably do not have the time in your day to review 75,000 data points. But with Sigma, with conversational AI, we're going to help provide that automation decision support tooling that allow organizations to really scale beyond
[11:18] what was previously a physical limitation in terms of the throughput of data they could generate um and again ultimately casting wider nets, making better decisions faster, um and accelerating the time that it takes to bring new products to market. Also, uh
[11:31] we're working on some agent agent workflows. As I mentioned, we have our own internal agents. Sigma has agents, you have your agents, too. Um the way that we see this future developing is into a a mix of hybrid workflows that include human actions uh physical lab
[11:45] actions as well as agentic actions. And that is why we're building a lab orchestration platform to help organizations kind of span all those different use cases. And that's really what's next for us.
[11:57] We're going to continue working with Sigma. We're going to continue developing the product. We're going to continue developing our scientific domain focused AI. Um and ultimately we're going to help organizations continue to evolve not just from fragmented landscapes to uh really
[12:11] strong analysis based landscapes but then continue on into really fully orchestrated automated actions and automation driven labs. And with that I will pass it back to Zach to talk a little bit more about what's coming up next with Sigma.
[12:23] >> Awesome. Thanks Michael. So everything that Michael talked about, this is like preAI, right? Like we we're still working with them to make sure that some of the newest capabilities that we are working on are also integrated in their
[12:35] workflows. And so let me kind of take you on a little bit of a journey of how kind of Sigma reached here, right? So we we started several years ago with primarily replacing like spreadsheet based workflows. You want to make sure
[12:47] that any of these kind of ad hoc like spreadsheet type explorations that you were doing in Excel, Google Sheets are all uh centralized now on top of your warehouse through Sigma. And so we that was our first kind of play. I think we
[12:59] we moved into more of like a app-based workflows where we introduced the functionality around input tables uh which allowed go right back back to data bricks. Um and then we started working
[13:12] on workflows and actions within these applications. And then more recently with the advent of like AI and everybody working on their own agents as you mentioned and and warehouses introducing a lot more interesting concepts around
[13:24] AI we're working on how kind of sigma agents and our native agents that are kind of helping you build and analyze things can al also help with your with your workflows. So we went from like spreadsheets about like four years ago
[13:37] to now fully agentic like applications that can not only just help you for internal operations but like luma it can help for your uh it can help provide value for your customers as well when you embed sigma into it. Um and so a
[13:51] little bit about our kind of platform we are very flexible. We have on top of uh a lot of cloud data warehouses including data bricks from a semantics perspective again we are not super opinionated. We inherit pretty much everything. uh if you have UC metrics
[14:04] like yes we will inherit them and convert them into data models and if you you can also bring your own choice of LLM again the common theme that you theme that you would see here is that we are extremely flexible based on your
[14:16] technical stack uh we inherit pretty much everything from the warehouse so security governance any any models or even agents right if you have genie spaces or supervisor agents great we don't want you to recreate them you can
[14:29] use them as pass through as part of sigma's experience or use them as tools in sigma agent Uh and then all of these kind of core AI capabilities help you again still um
[14:41] continue to build on the foundations of like spreadsheets, reporting, dashboards, rightbacks, apps and embedding as well. So if you think about everything that we've built, it is kind of stacked on on top of each other. And so none of the things that we've built
[14:53] are kind of throwaway because now AI is a thing. It actually works in conjunction with everything that we've built over the years. specifically on the AI side like this is kind of our our philosophy right so um
[15:05] as as Michael was mentioning there's like a vast amount of data that you need to do discovery on and so the first kind of pillar that we thought about is like okay well if the first thing that AI can make better for our customers is to be
[15:17] able to find things faster and better right once you find the right uh data sets or data models that you want uh how do we help you build things faster how do you get to the end point really faster is it a dashboard is it a report
[15:29] or an application or plugins, right? How do we help you build those things faster that Sigma already knows how to build? And then finally, this is where we truly differentiate because for the first two, I believe pretty much everybody on the
[15:41] floor will be claiming this. Um, when you see these insights, when you build these things, how do you take action on top of it, right? We want to close the loop. We want to finish the workflow. It's like when you see something that
[15:53] is, let's say, an anomaly, which I'm going to showing shortly, like what do you do after that? Like you send an email, you send an alert. Like so we sigma helps you kind of finish the flow and all of this is built on a very strong foundation of like enterprise
[16:05] scale governance and auditability. These are the things that we've worked on throughout the years. Um we worked with warehouses, we inherit all the governance and we have in incredible amount of logging as part of our product and you've probably seen a lot of that
[16:18] as well. >> It is it is quite something. >> Yeah. Um from a discover perspective I think the the first is like chatting a basic experience of like I want to chat with my data I want to bring uh all the
[16:30] warehouse integrations and I want to extend it through MCP u that's all possible already today and we're working with a lot of our customers along this lines building becomes even more interesting right because now you're going from a traditional BI experience
[16:44] of building just dashboard to full-blown applications you can ask it to create like survey applications headcount planning applications and it'll create a plan, you can work with it and make sure that you're building the right thing uh that you need for your end users. Uh and
[16:57] then soon uh we will also have code agent integrations. We are already live with cloud uh through our cloud connector but same with open AAI very shortly. Uh and then we are also introducing the sigma CLI. So again in
[17:11] term in the similar theme of flexibility we are not super opinionated about where you begin your journey with from right you could be anywhere um based on your tech stack and as long as you are able to connect to the sigma's MCP or use our
[17:24] CLI um through our agent skills you are able to create assets in Sigma which is on a very governed foundation which is data bricks um and then act right like
[17:36] conversational AI again soon probably going to get commoditized But these agents running autonomous workflows um really helping you sync your SAS. You might have seen our billboard here uh on one of the uh freeways where like tools
[17:50] that you tools SAS tools that are super opinionated about how they are built and um and created I think are probably going to go out of fashion. So here we are giving you the flexibility the Lego blocks that you can create your own u
[18:04] applications and your own journey and really drive action from into external systems because that's where really the the value is in finishing the workflow um and
[18:16] with action like let me talk about sigma agent I think this is where um a lot of our efforts are going u over the next few months. Sigma agents are part of these workbooks. Workbook is basically a construct where you can create any of
[18:29] these reports, dashboards, any of the applications. It allows you to set up instructions. Uh the the differentiating factor here is the dynamic component, right? So you can personalize these instructions per user. You can bring in
[18:41] the context of the application as you are building this agent. You can further govern and isolate the data that this agent need will have access to. So you can kind of prevent any kind of uh incorrect access for the end users. And
[18:54] then tools uh we have the warehouse tools uh like the genie spaces available. You have we have actions which is our internal framework available as tools and obviously you have MCP tools that you can bring in uh
[19:06] based on the additional context. In your example, if you have some specific like scientific context for a given customer in your MCP server, you can actually connect that as part of it and you can have again an oath kind of mechanism
[19:18] where your customer will only get context for them uh through >> I think that'll be one of the really exciting things for us as we connect RMCP to your to continue bringing that structure and that like second third degree layers of context that we have in
[19:31] our system um to really supplement the decision-m power of the agents. >> Exactly. And I'll I'll show you the output piece where you can really structure how the output should be. It could it could generate reports. It could send like you emails and so you
[19:43] can really uh create structured outputs but then you can have the conversation that you're having with agents convert into converted into like structured data as well that can be returned back to your warehouse and then you can use that
[19:55] for further analysis. So um really the LLM gets a lot of uh power on deciding like what what to do but Sigma executes it and but your governance is applied through and through through the entire
[20:07] experience. It follows the user agents won't be a a separate concept. It's going to inherit the permissions that a user has and so you can be you know you have one less thing to worry about from a security perspective. It's a it's not
[20:20] a net new concept. It just follows the user where it goes. uh from an architecture standpoint I already covered a little bit but we we uh we integrate with any external tools like Genie spaces it can you can bring like Google Drive or any other systems
[20:33] through MCP uh and through AI providers again we are very flexible you can bring in whatever is available in your organization uh uh that is approved by your security team including any warehouse based models that data bricks
[20:46] provides um so just kind of quick summary why sigma agent kind of makes sense in this world they are extremely governed like they have the same permissions that you have defined for the user in the warehouse. They are connected to your warehouse to your MCP server. You are
[20:59] basically creating this journey for your end users. They're extremely configurable. You control the entire experience end to end. And they're very actionoriented like a uh a user that is not super technical is able to configure
[21:11] an agent and have it running up and running in like a few few hours max. Um this this functionality has been in public beta for a few weeks now. uh and we have over a thousand customers who
[21:23] already tried and give us really good feedback. So if you're a Sigma customers, reach out to your your account rep. Um so let me show you it in action and um I'm going to attempt to do a live demo on conference internet. So
[21:36] let's let's see how that goes. Um so Michael was kind enough to send uh some bioreactor gas data set. Uh how would you explain like what this data set is and I I'm going to ask a question about
[21:48] finding some anomalies. Um, I know nothing about the science behind this. I'm just going to pick O2 and say, tell me anomalies in O2. >> Yeah. And so, while the agent's thinking, right, for those who aren't in
[22:00] life science, um, basically a bioreactor is a vessel that allows a substance to grow. So, that could be like a cell culture that's come from a parent batch that you're going to be using for downstream testing. It could be a
[22:12] biologic that you're actually trying to grow that is the therapeutic itself. But these things are really important because depending on what it is that you're using as that seed compound, you could have anywhere from hundreds to thousands, tens of thousands of dollars
[22:25] in reagents going into the reactor and trying to grow it. So what this data set is is effectively monitoring the parameters of the bioreactor over time. How much oxygen's in the chamber? How much carbon dioxide's in the chamber
[22:38] etc. So what the agent's doing for us right now is looking at that and saying okay if I know that this thing is being used to govern to to grow something and this is not context that Zach had to add itself. It just is something that it can
[22:51] know by the virtue of the fact that this is a bioreactor. It's basically looking at those parameters and saying okay like is this run okay or not? And now that the agent's kind of giving back some results I'll I'll pause there uh for
[23:04] Zach to talk through this and I'll kind of talk about some of the impact of this. But as you can see with minimal context, we're already starting to get back some really amazing and powerful results. >> Yeah. Again, not a scientist, but I'm
[23:16] going to scroll up and show you. So, it found um so what it what it did is like it created kind of these like mini tables as it's kind of finding context and identifying what data is relevant for it to analyze. Uh and because we are
[23:28] an overachiever, we'll go beyond just the O2 and help figure out what are some of the related things uh that this agent this agent has found. And so first it found the the anomalies that I asked it to do. So critical O2 anomalies and I'm
[23:40] I'm assuming these are like uh reactor numbers. >> Those are probably yeah reactor numbers or uh sample identifiers. One of the two. >> Got it. Do you think this is the right analysis? Uh just like eyeballing it.
[23:52] >> Yeah, absolutely. This this is the kind of thing that I expected it to basically look to to say this this is what the critical findings are. These are instances where the run is either potentially going at risk um or has
[24:04] potentially gotten to a place that that it's going to fail, >> right? >> Um just just like us, just like any organic compound or most organic compounds outside of DC compounds, right? Deep sea. Um you know, things need oxygen to survive. If that starts
[24:17] going out of whack, somebody needs to get to the bioreactor as quickly as they can. Otherwise, you could lose a run. That's tens of thousands of dollars potentially in reagents and seed compounds, plus the time. um the the
[24:30] physical time that it takes to actually seed these runs allow things to develop to the point that you can use them downstream. I mean it is a huge and real cost. So to be able to use the agent like this to actually have this process
[24:42] automate and passively monitor the systems to send alarms and notifications to people to get to the instrument and act on it before you lose that run um is a huge ROI for companies. >> So I think the last part you said is very important because anybody can point
[24:55] an agent to your data and like ask questions. I think it found really good things because obviously our models are great but I think the interesting part is u not having to come to this workbook or a dashboard or an app all the time when these things happen because when
[25:08] you find anomalies you just need to be notified rather than having to go to something and ask if there's an anomaly which is what um we kind of have done right and so beyond just asking questions off of your data we also want
[25:21] to say if you find anomaly send an email to me maybe check check every 5 minutes 10 minutes depending on you know how how stable it needs to be across time uh and then generate a full report and send it to me right so it basically that's what
[25:34] it opened up on the screen where it created a full report based on everything that it found uh in a very structured format this is also a plug-in that we've created for like an email um PDF uh and then through our agent we can
[25:48] actually make it take an action so let me just go under the hood here for a second and then I can show you an action sequence that we've created And so it basically found these anomalies, inserted this
[26:00] information into like an input table, wrote it back to data bricks and then converted that information into a very structured formatted PDF and then it will send you an email um every time it finds anomaly. So again it goes from
[26:14] discovering um your data to asking questions of your data and then taking an action on it right away. You can also plug in some API calls here. Um we have a if you look at a series of our actions
[26:27] that we support you can make API calls too based on any u action that you might want to take. So the agent uh can actually find these anomalies uh send an email to you bring it will
[26:39] bring you back to the to this uh application and then you can take further action on it like okay if you have some APIs to stop something or or whatever recommendation it has provided in this it can actually do it on your
[26:51] behalf too. Yeah. And I think the other thing that's exciting is you start to think beyond this, right? Everything that Zach mentioned is structured. It gets committed to the database, which means now you are also starting to feed that back into the experimental record.
[27:03] And the structured approach we take in Luma to say, okay, so look at the trends across these hundreds, thousands of bioreactor runs. Are there specific parameters that we need to avoid from an experimental design perspective? And ultimately it just creates this really
[27:15] self-feeding feedback loop that really helps organizations mature and kind of lose lose less information and less runs in the future. >> Totally. Yeah. So now that we're on screen, let me talk a little bit about like this agent. This UI what it is,
[27:27] right? So this is an agent configuration UI. On the left side you have a agent that helps you build the agent. Uh it's an inception uh of agents basically. So I I I actually did not write all these
[27:40] things myself. I just came in and I gave it some very basic information uh about the data set. I so I I connected my data sources. I connected my anomaly log that I wanted this agent to write to and then I gave it some very basic instruction
[27:53] about hey like uh this is a you are a bioreactor monitor and this is the data you have access to and help me find anomalies and it created this massive instruction set based on this very basic information I provided. Um and then
[28:06] through this natural language uh interface you can also create these actions uh and action sequences and you can say okay every time you find anomalies insert rows into this table export it by email and then navigate to
[28:20] uh this report that I just created. I don't know if my email is connected but I can also show that this email actually went to me um as soon as I it found some anomalies. The interesting part here is this governance, right? And you don't
[28:33] want agents just taking actions because they can. You can also add human approvals here. And so right now I did not have any approvals required. So it send me an email immediately. But I can if I say it requires approval, it will ask me if I should actually take this
[28:45] action or not. And so uh you have governance at really every layer where the data is governed uh from the data from the data warehouse. So data bricks security gets inherited. you have u
[28:57] another layer of governance through like Sigma's permissioning layer. This workbook has its own grants and then even within this agent you can confine the resources uh what this agent has access to and for each action you can
[29:10] actually have an additional approval to make sure that the agent doesn't take any action uh without your permission.
[29:23] Cool. Um let me switch back to my slides. All right. So that's a little bit about the the bioreactor demo that we did. Um
[29:35] just a little bit talk about enterprise scale governance and auditability. Right. So data bricks obviously like a power powerhouse. Uh it has all these applications uh that we built
[29:47] they're built on live data that is coming directly from data bricks. There's no movement. We don't really take the data out of data bricks to to make these apps and workflows possible. We run on on live data. Uh and the all
[29:59] of your governance that you have set up is all intact. With Sigma, we bring in really an enterprise ready solution where we inherit everything and instead of wipe coding things in an environment that is not governed like it provides
[30:12] you like a nice little sandbox to to experiment, to play with, but then also create production ready applications. Um apart from dogmatics we also have some of the other uh customers in this space
[30:24] um who have who have like interesting challenges that they have solved using Sigma's capabilities with HSS we have the real-time financial visibility use case that we we solved for them and the with uni liver it was around supply
[30:37] chain command center um and happy to talk about these in detail at our booth number 106 um and truly this year has been a year of innovation for both data bricks and sigma and we've we've
[30:50] partnered with them on a lot of these initiatives that together and I think that's you're seeing the benefits of it through that um we've we have introduced a sigma assistant that runs natively on any data bricks AI models that you
[31:03] connect uh it every time you upgrade it gives you a higher quality we have AIdriven workflows so the genie spaces can actually be integrated as part of tools within the within the interface uh
[31:16] and then you have the enablement of any kind of AI powered application through our builder where you can generate all these applications just uh through natural language. Um and then from an advanced analytics
[31:28] uh perspective we have Python for data bricks now available in Sigma and really notebook style workflows where you can use this uh Python element on live data. Uh what's coming AI powered analysis I
[31:40] think we have barely scratched the surface of what what's capable uh what Sigma is capable today. Um and we are figuring out what's next for us, right? And so the the mental model of asking a
[31:54] question, getting an answer on static assets like that makes sense, but we are also figuring out what a a different set of mental model where you can u you can start with pretty much nothing and you
[32:06] can bring in data as you need and generate artifacts uh as you're having conversations rather than going to static assets and then asking questions off of them. uh from an AIdriven workflows perspective again I think that actions framework is just the beginning
[32:19] of what we want to do uh the interconnectivity of the agents basically how our agents can work with your agents uh directly either through APIs through an MCP um we are also exploring that in the future and then
[32:31] really advanced analytics we want to simplify it right like we are the second time BI partner of the year with data bricks we want to continue advancing on on on this path we want to make sure that every innovation that data bricks
[32:43] is doing we're able to um kind of express it better through Sigma's interface. Uh and that's where a lot of our efforts are going for the next you know six or nine months. And uh that's our talk. Thank you.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.