How J&J MedTech and Takeda Scale Data & AI

Summary

J&J MedTech is using the Databricks Data and AI platform to de-identify surgical video at scale, govern lineage in Unity Catalog, and share datasets via Delta Sharing, building a multimodal surgical AI foundation that bridges scene understanding and situational awareness in the operating room.
Takeda Pharmaceuticals has invested significantly in a centralized governed lakehouse on Databricks spanning commercial, research and development, and manufacturing, with agentic workflows that have reduced pharmacovigilance submission trough workload by approximately 30 percent.
Both companies represent a broader shift in life sciences AI—from descriptive analytics to generative science—enabled by a governed lakehouse architecture that unifies structured and unstructured data across GxP and non-GxP environments.

How J&J MedTech and Takeda Scale Data & AI

this video features thought leaders from Takeda and J&J MedTech on building governed AI foundations. Takeda shows how pharma AI stalls on disconnected data governance, not model quality. Learn how they're extending one governed Lakehouse on Databricks across commercial, R&D, and manufacturing—turning descriptive analytics into generative science across generative chemistry, adaptive trial design, digital twins, and agentic workflows where the platform is the advantage. Surgery is high-stakes and deeply personal. J&J MedTech will then explore how the OR's richest data—surgical video, room video, and robotics—is unstructured, massive in scale, and deeply sensitive. Learn how their AI and engineering team uses Databricks to de-identify surgical video at scale, govern lineage in Unity Catalog, and share datasets via Delta Sharing—unlocking a new generation of multimodal surgical AI that bridges scene understanding and situational awareness in the OR.
Talk By: Christina Busmalis, Global GTM Leadaer, Life Sciences, Databricks ; Daniel Carchedi, Global Head Partnerships & Alliances, Johnson & Johnson ; Nikhil Gulati, Johnson & Johnson ; Raman Singh, Head of Enterprise Data Platforms & Srvs, Takeda Pharmaceuticals - USA ; Shyam J Dadala, AI & Data Platforms Arch and Strategy, Takeda Pharmaceuticals ;

Chapters

FAQs

How is J&J MedTech using AI in the operating room?

J&J MedTech is building surgical AI capabilities including anatomical detection, OR optimization, and automated post-operative notes, using surgical video, room video, and robotics data as inputs. This video explains that J&J uses the Databricks Data and AI platform to de-identify surgical video at scale, govern lineage in Unity Catalog, and share datasets via Delta Sharing to train multimodal models.

What is Takeda's governed lakehouse architecture for pharma AI?

Takeda built a centralized lakehouse on the Databricks Data and AI platform that spans commercial, R&D, and manufacturing divisions, including GxP environments. According to this video, the architecture supports generative chemistry, adaptive trial design, digital twins, and agentic workflows, with all data governed through a unified platform rather than siloed departmental systems.

What are pharmacovigilance agentic workflows and what impact did Takeda see?

Pharmacovigilance agentic workflows automate the review and submission processes required for drug safety monitoring and regulatory reporting. This video states that Takeda's agentic workflows reduced pharmacovigilance submission trough workload by approximately 30 percent, freeing safety scientists to focus on higher-value analysis rather than manual data processing.

What is the 'digital surgical flywheel' concept described in this video?

The digital surgical flywheel describes the infrastructure required to turn raw surgical data into a continuously improving AI system—capturing multimodal data from the OR, governing it on a secure platform, training models that understand surgical scenes and surgeon performance, and feeding insights back into clinical workflows. This video identifies key gaps that must be solved to close the flywheel loop in real-world hospital environments.

Full transcript

[00:08] Daniel Cartetti and I'm here with my colleague Nuel Galotti and today we're going to talk a little bit about polyonic but also uh the o tell you a little bit about the operating room and we'll also talk a little bit about how J&J is working with data bricks to to
[00:21] partner on accelerating surgical innovation. So let's start with the O. I'll get us started with the O O really is a special place for those who haven't had an opportunity uh as an observer,
[00:33] not as a patient to be and it's really obviously where things like cancers are removed and joints are replaced and and lives are saved. uh but it there's a lot of procedures that happen globally to the tune of at least here in the US 15
[00:45] million procedures happen globally about 310 million procedures happen and what's happening in the O is it's really bringing together uh highly skilled clinicians in a remarkable intersection
[00:57] of human expertise but also operational complexity and cutting edge technology and so we believe uh the O is really going to become even more magical and
[01:10] special. As we move forward, we're at a real transformative change time period for the O. There's a rapid evolution and advance of visualization tools. Um, surgical robotics have really gained
[01:23] mainstream acceptance now. And as you would expect, AI and the role it plays uh is becoming more and more prevalent in the O. Um, fun fact, last year uh 2025, um, the FDA, it was a breakout
[01:37] year for the FDA. there were almost 400 AI or ML enabled device software functions approved for for the O. So a lot a lot is happening and we're witnessing a shift that's happening a
[01:50] shift really historically as you would imagine where uh a surgeon a solo surgeon he or she would be in the operating room and have their expertise to now moving more to a shared intelligence environment. And so let me
[02:04] draw a quick analogy that it's probably often overused, but I think it would be relevant to what we're talking about today. Um, when I came here from the the airport, I Ubered. Um, obviously that is
[02:16] tapping into collective knowledge of data that millions of drivers were receiving that real-time updates in terms of the hazards on the road, the traffic, uh, adapting to those traffic conditions. um making our journey not
[02:30] only more efficient but obviously safer for all of us uh the passengers as well as the other drivers. The same shift is happening in surgery. Um while today's o kind of relies on these individual
[02:42] experts, we're seeing that shift happen where clinicians are going to tap into really a global cool uh pool of collective wisdom. And so AI systems are now learning, they'll continue to learn
[02:55] from millions of procedures worldwide. um will identify patterns and really insights that really no single surgeon could he or she uh see on their own and that's where this data that we see for
[03:09] making self-driving cars um smarter that data and AI and surgery is also going to in elevate the surgical experience um and also empower every surgical team to
[03:21] access best practices right and to reduce variability when it comes to operations and ultimately improve safety and for every patient. So, let me uh dimensionalize this shift
[03:34] with a few examples and I'm given a little bit of viewer discretion because if anybody has never seen laparoscopic video, you might want to turn away or might want to close your eyes. Um, this is a laparoscopic procedure. So imagine
[03:47] if we could identify critical anatomical structures in real time, ones that tend to be invisible to the human eye, whether it's nerves or vascular structures. We could visualize no go
[03:59] zones for the the instruments that you're using in in this laparoscopic procedure. We could we could really gently guide surgeons throughout the procedure, offering them the eyes of thousands of experts in the field. we
[04:12] would be able to really essentially democratize surgical skill and drive towards better surgical outcomes and these technologies and this visualization software exists today.
[04:24] Imagine if you could uh understanding what the surgical scene enables. We could really reduce the crippling administrative burden that's happening on all surgeons across the world by automatically generating post-operative
[04:38] notes. So taking ambient listening from the surgical theater and really then taking them creating post-operative notes, mapping them to insurance codes for example. This really could help with
[04:50] physician and surgeon burnout. Allow surgeons to focus on what brought them to the profession in the first place and that is that's patient care. We could dh dynamically predict delays
[05:03] in the O. So whether a procedure operate, you know, the operating room itself to be able to uh optimize the O scheduling, turnover, staffing, hospitals in in hospitals right now
[05:16] getting the most out of arguably their their most expensive uh valuable asset and that's the people in the OS. Just to give you context, today's fully loaded O costs uh in the US about $100 a minute.
[05:29] And so productivity in the R is measured through case throughput. So how how efficiently are you using that operating room? It's critical for hospitals uh financial viability even really small
[05:42] negative imbalances in productivity could actually u lead to a health system being insolvent. And so these are just a few examples of how AI is transforming
[05:54] surgical practice and will continue uh in the coming coming years. But today let's let's talk about the reality of today. Right? So for for those who don't spend time in the O, we're we're not quite there yet. I mean
[06:07] not at least not at scale. So some of the things that I shared with you are available as point solutions. Um and really the stakes remain incredibly high. We know that across surgeries globally. Uh mortality rates still
[06:19] hovered about 3% 3 to 5% in 2026. Um we know that surgical complications typically occur one in three patients. So some of them are severe, some of them require u readmission and some of them
[06:32] are clearly life-changing or costly. And when it comes to people uh the surgeons themselves as I mentioned earlier uh burnout is a big issue right um we've made some progress post pandemic uh but
[06:46] still almost half of US physicians report at least one symptom of burnout. The data on surgeons is even higher. 40% of surgeons surveyed recently uh consider leaving the profession or
[06:58] retiring early. So this is a huge issue. 25% of practicing surgeons in the United States right now are over the age of 65. And so the American College of Surgeons
[07:10] reports that um and predicts that basically they'll within the next 10 years they're going to age out and retire. And so we'll have a shortage of close to 10 to 20,000 uh surgeons uh within the next decade. And then the
[07:23] last thing is really the O as I mentioned earlier uh it's really the economic engine for a hospital. So 60 to 70% of hospital's profitability comes and revenue comes through the O. So
[07:36] making that more efficient is critically important from an ROI standpoint not only for patient care but also for the hospital's viability and profitability. So there's good news here I guess in the
[07:48] sense that for the first time in history technology is no longer the bottleneck here um to solve these thorny clinical and operational problems. We're we are creating a ton of longitudinal
[08:02] multimodal surgical data um predictive and genai allows us to turn that clearly that data into insight and action. um big tech is really knocking down the doors to to beg for use cases to really
[08:15] implement this AI infrastructure. But more importantly, there's a big push right now in institutional momentum. So what's happening now is um these these institutions, these hospital systems,
[08:27] they recognize that they need to leverage this technology for the reasons I mentioned. Uh and so we're seeing the shift that's happening. So the long story short here is to unlock this next wave of innovation is
[08:39] we really need to get you know what we call this digital surgical uh flywheel going um to really innovate in the O and doing so at scale we need to do it and find ways that uh address these infrastructure gaps. So the the flywheel
[08:54] that I'm sharing with you here uh won't start spinning until we're able to um unify multimodal data. we're able to synchronize these data sets uh in a compliant way. Uh we also have to deal
[09:07] globally here. So there's different uh global constraints. Um we need to reduce the effort to get AI surgical data sets and accelerate model development in a secure environment, a trusted environment. And until we unleash these
[09:21] these uh models and deploy them seamlessly uh into hospitals, we need to do it in a workflow that they're used to and comfortable with. Uh and then the last thing I'd say is we need to make the value creation we need to create
[09:35] monetization opportunities for companies um and create this kind of fair value creation. So um for all the contributing stakeholders here. So at J&J uh we are building a platform and an ecosystem
[09:48] called polyonic that we believe is a path forward to achieve this. But to do this, we we clearly have technical challenges that many of you in the room deal with with other use cases that I'm going to turn it over to my colleague
[10:01] Nail to talk to you about how we're approaching it and how we're addressing it from a technical standpoint in partnership with Data Bricks. So Nail, >> thanks Daniel. >> So let's uh dive right in. How do we
[10:15] actually make that uh flywheel work in in service of our patients and uh and the surgeons? uh I mean we start with uh talking about multimodel data. This is a data and AI conference and we have
[10:27] always already heard that AI does not have a uh data problem. It has a context problem and the context from the O is built when you actually combine different uh streams that are coming from this O right and and and what does
[10:40] it mean uh for someone who's actually not seen uh what is the representative data in the O. So let's talk about video. Video is what's what tells us what is actually happening in the O. So
[10:52] if you see minimally invasive surgery or robotic assisted surgery, uh the eyes for the surgeon is really the video that is actually capturing an example that Daniel showed what's happening inside the body. So that gives gives us really
[11:04] the what um and then we have kinematic data and I'll talk about each of them in the next few slides. Kinematics is really uh more specific for robotic assisted surgery. So it's projected that in by 2030 we will have 15% of world
[11:19] surgeries done by robotic assisted surgery. Right. And that means that robot is actually an integral part of the surgery and also the extension to do the surgeon. Uh kinematics data is really encoding every single movement that surgeon is is doing. Uh and that
[11:33] data is so valuable because it actually tells you uh to differentiate between skills of the surgeons, what is actually happening on the device, whether it's reliability or or or just the operations
[11:45] of it. And eventually there's a lot of research that tells us that that can be correlated with patient outcomes, right? That could actually differentiate another uh layer of data that has not been available or locked pretty much in
[11:57] within the OEM ecosystems to tell you what happened to that patient 30 60 90 days because what happened within the surgery today we don't we don't actually get to do that analysis and telemetry
[12:09] same thing. It gives you the operations of the data and now we're also familiar with things that have been around for decades actually. Ehr data but no one actually tries to combine all of these together and create a full view or a longitudinal view of uh what happened in
[12:22] that surgery and what could happen to patient outcomes down the line. Right. So back to video. Video is really a rich source. Uh it it it is actually a dense source. So extracting context from this
[12:34] video is extremely challenging. You need domain expertise. You need a lot of annotations. You need experts to tell what is right, what is wrong, what is the ground truth. So we we spend a lot of time thinking about that uh how do we actually digitize this? How do we use
[12:47] some AI capabilities and and and uh in our platform to actually extract that context and then give it to our downstream applications as I said about kinematics you know it's really it's a very it's a toy example
[13:00] but think of it this as a signature of a of a surgeon. So how a surgeon operates a a robotic device, you can actually create a very unique uh signature for that operation and for that surgeon and
[13:13] eventually if you collect enough enough of this, you kind of starts to see what best looks like and you start to create benchmarks from it and and you know there's a lot of research there. So we we we like to believe that that research can be productized into workflows and
[13:26] that actually unlocks some of these uh uh use cases. uh but let's talk about today and what we are doing about this. So five to six year years ago if you if you talk to a
[13:38] lot of these institutions and and surgeons they would tell you we we all need this right but the reality was the d all of this data or most of this data was locked within within the hospital systems uh you had to really figure out
[13:51] a way to safely compliantly pull this out uh because there's patient data involved there's PII involved so there's a lot of these challenges which are not necessarily technology challenges there's compliance infrastructure that
[14:03] needs to be put in place. So we really think about left to right what's happening in the O what's happening in the hospital and how do we pull that data first into a system into a cloud into a platform that we can process it at scale right so we we actually put an
[14:17] edge open edge device in between which actually can connect to different data sources we connect to different OEMs it's not just about whether we can serve uh of course our own devices but also
[14:29] devices that are not J&J devices within the O how do we create that ecosystem to plug into it. But once that's there and we've been spending a lot of time thinking about how does that work work with our hospital teams to do that, it
[14:41] really lands into this this sort of like a right side of this picture where we are leveraging a lot of database capabilities to process the data at scale uh but most importantly put a governance layer before everything
[14:53] starts right. So think about um every producer consumer of these data sets u all the way all the way to the row level uh who asked for what what application consumed it what processing step was put
[15:06] on each of these datas where the deidentification happened all of that is actually managed within our um UC uni unity catalog so if if you're familiar with that uh that builds a governance layer for us and then once we actually
[15:20] enforce that compliance and access policies or even contractual encoding of the policies in this system. Uh we can we build surgical intelligence layer on top of that because we have solved a hard problem there and then the
[15:33] intelligence layer really accelerates these workflows for building these models for data scientists engineers. So if if there are 10 steps to finish a you know critical view of safety model we've already done seven of them for for that
[15:45] consumer. So they can actually accelerate uh using pre-built ML pipelines within data bricks. we can actually deploy them through uh you know the whole goal is to go from development experimentation to production really
[15:57] fast right so all of this uh stack gives us all these capabilities to accelerate that once we solve the compliance and the data ingestion problem the other thing that I want to highlight is uh because we have this ecosystem approach
[16:10] we really think about uh third party data exchange and and producers and consumers so technologies like delta sharing and sitting on a common format like iceberg allows us to actually do that right off the bat and we obviously
[16:23] work with rest of the ecosystem partners to make sure that you know they can follow those standards and then we can we can create that exchange seamlessly. Um there's a lot going on this slide uh
[16:35] and uh that's intentional because I wanted to actually uh showcase there's a little there's a lot going on in the O right if you really wanted to tap everything that's going in the O you got
[16:47] to start thinking about everything that's available for you to you to do not every company does have this view either some folks look at the device or some people look at uh non-rootics devices or some people look at what's happening in the O we want to actually
[17:00] look at uh everything that's happening because that's the way to bridge uh what's happening in in the war which is what we call situational awareness and what's happening inside the body when the surgery is happening. So that's what
[17:12] we call surgical scene understanding. Uh a couple of examples would be you see there's a there's a team there operating and if we can figure out the movement within the O we can figure out who's doing what most efficiently or not and
[17:26] then but we can correlate that back what's happening during the surgery within the endoscope video link it back to some of the EHR information then you start to really do three things one you actually can uh understand what was done
[17:38] but you can also calculate insights and automations for from that second thing that what you do is you give both the surgeons, the hospital administration and patients a single view of what happened. So there is no debate like you
[17:51] can you can all come together a lot of times we have talk spoken to surgeons and they'll tell you I actually reviewed the surgery back with the patient because now I have a video so I can actually show them what happened at you know you know minute 20 or what happened
[18:04] to you when we were making these decisions. So now you have a single view that can be shared by multiple parties if we can able we can collect that data and actually show it in that way that eventually starts to create those five
[18:17] right then you have the use cases then you have the buyin from these stakeholders and that allows you to now say these are the most important use cases I got the data there's there's a there's a real need for it and we can
[18:30] actually deploy that back into the same ecosystem and this whole thing runs again and again right and and I think in the end. Our goal is once we have enhanced situational awareness and a real semantic representation of that of
[18:42] that journey, we could in my personal belief achieve this this sort of holy grail idea that every time you walk into the surgery or you're a patient or a surgeon, the ambiguity and and what
[18:56] happens down the line to that surgeon from a recovery perspective, you reduce that ambiguity more and more. Thank you. >> Yeah. The last the Thank you. The last thing I leave you with, we'll just leave
[19:09] you with the fact that I think the that the O uh room of the future is not a room, it's a learning system. And I think what we're doing with Polyonic is as uh Nicole mentioned, it's it's an open device agnostic ecosystem. Um and
[19:23] it's a unified experience. So if anybody's interested in learning more about what we're doing or interested in partnering with, let us know. So thank you and we'll turn it over to our colleagues. >> So thank you very much Daniel and Nikki. Fantastic. And it's a great example again of the future of health of being
[19:36] interconnected intelligent and personalized really driving that that benefit within the operating room. We're going to continue on the journey in life sciences and pivot now toward towards biioarma and glad to have that teada on the stage
[19:48] with Sham and and Ramen who's going to talk about really molecule to market. You've probably heard that phrase before um and their journey that they're with us within the data. Ramen over to you. Thank you. >> Thanks China.
[20:01] >> So this is about the tea. Uh I'm Rahman Singh. I head the enterprise platform ops and I have my colleague uh Sham. He he's the head of uh data platform architecture and technology. Uh so we'll walk you through a journey um from where
[20:15] we were and where we want to go in terms of biioarma in terms of our farmer value chain. So this is a little about tea who we are and let me just give you a little glossery of uh essentially you know
[20:27] where we stand today right we are we were founded uh 240 years back uh headquartered in Japan the global hub for tech innovation headquarters in Cambridge around 50,000 employees um we
[20:39] have almost uh you know 28 billion of revenue and we spend almost $5 billion on R&D so as compared to top 20 I think we are still like on the upper percentile of R&D spend
[20:51] We have six late stage pipelines uh two this year and next one year. So very rich pipeline and we have seen the journey uh in terms of how we were using datab bricks as a core platform across
[21:04] the ecosystem. The next slide I'm going to walk you through it gives you the breadth of our uh datab bricks adoption. So we were one of the earliest adopter of data bricks. In last 18 months if you can see we have
[21:16] almost 9,000 active users right 12 terabyte of data is processed per month 307,000 jobs 20,000k pipelines and and so forth right uh we spend almost 70
[21:29] million on DBUS in last 18 months so that gives you the breadth and scale of a real data foundation we have built the foundation is gone it's is it's it's already at scale so then we are saying
[21:43] why do we How do you apply science now? Why now? Right? That's a question we asking ourselves, right? The investments we have made in partnership with data bricks, they've been a great partner so far. How we can scale this up to drive the value in the farmer ecosystem?
[21:56] Because as a head of the platform team, I need to justify where we want to go in terms of science in terms of a genic AI that's a question I think some of you would be being asked. We've been talking about token max right across the board. last two days I'm hearing about the
[22:10] token max has no meaning if you are driving the value across the organization so the value from pharma comes not from the corporate reporting or manufacturing it really comes in the R&D in science and how can commercialize
[22:23] those launches for the drugs that leads to the next slide I have it's a kind of a generic slide and that I think is very data from McKenzie every drug cost $2.6 6 billion dollar to launch
[22:36] and 12% of the candidates never go beyond phase one. The AI price of pharma is between 60 to $110 billion huge and this data probably is a little old it probably much higher with aentic AI AI could uplift the RD
[22:51] productivity by 10 to 15%. of the total cost but 95% of what goes has no value that's a big big important part 95% what goes into the effort of AI has no value
[23:04] I'm talking about the farmer ecosystem and the reason is not because technology doesn't exist technology evolve rapidly is because people don't have a value I think you mentioned context we hearing from Ali yesterday contextual use of AI
[23:17] across a board is super important that's why I think we using plan to use and in the process of using it the platform to find integration governance and quality data across the layer integrating a
[23:30] platform with R&D platforms. So we use Tetra science heavily now how to use data bricks in conjunction with those investments and drive the value value across the farmer value chain. So what brings us to pivot? This give a slice of
[23:43] I think most of you can connect most of us use our data bricks platform. I'm not talking about technology with dashboards mostly retrospectives operational commercial reporting data that test
[23:55] which is after the fact it's mostly reactive. We want to pivot the same platform same investments towards more generative science more predictive analysis.
[24:09] So critical point is we don't want to build a second platform a new layer right same platform same process governance but the new mission that new mission is to take the current investments we have and compound across the new workloads across the whole
[24:22] science and R&D ecosystem this is a farmer value chain where I think you can connect from the standpoint of research and development preclinical
[24:34] submission commercial and market access to to marketing and the post-launch there are a lot of value drivers where AI can help us out. So from the standpoint of drug discovery and then again this is not
[24:47] just data bricks we have partnership with multi multiple companies called lampic bio we have partnership with nebula and we can talk about other partnership whether it's through nvidia this is all of the above right but I the
[25:00] AI is a common factor precon preclinical specifically I think is adding on on the stage zero to stage one whether we use insilico that's what
[25:12] we're doing now with the lab of the future, right? But the biggest bang for the buck, I'll show you my next slide, lies on the commercial launch, the uplift we get on the revenue using the agentai workflows on the commercial launch, right? From the standpoint of
[25:25] medical affairs branding to go and having the adoption of the drug across the SCPs is super important. So this slide is very super critical slides. We have done analysis if if you see that
[25:37] this trough happens when we go for the submission. So if you use again these a benchmark um I mean uh depending on which way you are in your maturity on a process side this will be different curve but the value is
[25:51] huge especially the value of death comes with the point of submission we spend the money across a board if we use the agentic workflows we'll have a up to 30% shallower trough but then as we move for
[26:06] forward the AI earlier zero crossing happens much earlier Right. So what analysis we did every delay in the launch cost $5.9 million for for drug that's a typical benchmark with McKenzie
[26:19] with 10 different analysis we did uh using the obviously AI. So if we can use agenting enabled workflows we talking about huge savings and huge uplift to the revenue. The peak uplift happens on
[26:33] the step 29 of the pharma value chain when we have market launch commercialization of a product. Right? So the benchmark is for based on last top 10 10 selling drugs 1.2 billion
[26:45] above the baseline of the peak. So if you use traditional workflows, traditional process and then you overlay with the AIdriven workflows across the chain and some of the workflows I think there still FDA is still not not up to
[26:58] the mark but if we go to the the future state we talking about billions of savings and billions of potential revenue to the farmer company.
[27:12] So so where we are focusing first what's important drivers why we doing this let me take to the next slide. where AI meets the science. So I've been talking about the the the the right from uh stage one stage two to launch
[27:25] commercial. The foundation you see in the bottom is the data platform one platform which is interoperable. This is the database we use and every every organization has multiple platforms. So you cannot derive the
[27:39] value across uh the farmer value chain with AI without having a unified data foundation. I mean you heard about in last two days from Ali from various leaders data is the key clean data cured
[27:52] data uh not duplicated data whether you're buying data from say inqia or third party it has to sit on one foundation every R&D has multiple data warehouses right how to build that
[28:04] unified layer is super important now with ontology and everything coming on I think that's a game changer so once you have that ontology layer the relationships are built semant antically
[28:16] right that can drive the value to next notch. So I'm talking about before the announcement I heard two days back uh with that layer I think it's a game changanger for most organization.
[28:31] So I'll pass over this to Sham to talk about a little bit about the data platforms we have and the architecture we build to take us forward. Come sham. >> Thanks Raman. >> Hello everyone. Uh before we say where we stand today, right? I want to talk
[28:44] about where we were couple of years back uh similar to every global company we had same challenges with uh data silos as um I want to kind of index on what Ali was mentioning yesterday uh when
[28:58] he's talking about the game of throws like you have data silos. We started a journey for the data silos and we try try to merge everything into like making one unified platform that is where we can govern when I say platform it's not
[29:10] just data bricks it's a ecosystem that we build building to make it more uh a ready uh govern well with all the policies and whatnot. uh the four layers the the thinking from the user
[29:22] experience to the top down approach right we created four layers one is user experience layer where users can exper have the same similar experience where you access the data through apps or data
[29:34] bricks dashboards or whatnot and uh the to enable the journey that we are trying to embark into innovation and science that is where we are enabling AML framework through databicks mosiac where
[29:47] you can enable data science deep learning AI I ML and other things. So the core of any platform is data. So that's where uh we had a lot of silos. Now we trying to move into more feder u
[30:00] uh data mesh where um we're trying to align all the data sets with um uh domain construct that is where we harnessing the elm limbs data to make it a ready and everything is running on
[30:13] like the lake platform. we we centralized we have uh GXP works in a separate way and uh R&D is going to be it having its own uh lakehouses but end of day because it's a one unified
[30:26] platform we trying to uh govern through one place where you have proper lineage we have governance we have metadata and uh enrichment through ontology that we learned recently and cross cutting the
[30:39] governance is through unity cat that we enable So, so this is where I want to uh uh uh zoom out a bit. The four layers from the experience layer we are looking at. We
[30:52] are not not trying to change the experience because some of the users where in taka they don't want to change the experience that's where we we want to keep it like similar experience digital channels or apps or uh APIs or teams or wise and uh the AI at core this
[31:07] is where we are enabling uh both traditional uh data science as well as the agents that coming out of the hyperscalers that we use today. Primarily we built on Mosiac data bricks
[31:19] framework but uh like as in we when we expanding the AI foundation we are also bring trying to bring the agents coming out of the box from the other hyperscalers. So the bigger uh buck is
[31:31] on the zero data mesh that we're trying to create. not trying to move every data into one silo but this is where the mesh helps you bring the data from different pockets of taka R&D manufacturing plasma or uh corporate um everything like the
[31:46] data and AI control toward that monitors your data governance AI governance ethics compliance everything that because that's core of any AI uh success story um adoption we are enabling
[32:00] through like commit practices learning sessions and whatnot but uh end of end of the day like the foundation is built data bricks as a primarily core platform and uh we enable like last year we
[32:14] started this journey and uh if you want to learn more about a foundation is doing one more session tomorrow we tune into session tomorrow with that I'll hand it over to Raman
[32:28] >> thank you all right so I have few more slides and I think uh timing wise is no good right so this is I think important slide um I'm coming back to the theme unified data platform
[32:41] is advantage right so if you have data sitting in silos every organization is very fated at least farmer right I know right there are different BUBFs having a you can use any architecture data mesh architecture fabrics or you can put any
[32:54] architecture paradigm right having a one common view of data and quad data is super important right So right now if you see the number one step the tikita is already in motion uh since we early adopted data bricks maybe six years back
[33:07] with sh and then onto the the corporate uh R&D I think we are now integrating slowly right and it's always the case uh we do want to touch R&D because that's a DNA of farmer so we in process of
[33:19] ensuring that u the dark data which is like drug data sitting over there for previous trials we have access to and and we have strong partnership with our head of R&D a team and we're collaborating Um and to have a unified
[33:31] view of data is important. So our AI platform is already in production. The process are deployed. U you know the the geni agents I think we've built a lot of investment last two years. So the pivot number two is in progress right now
[33:44] which is towards more science uh from from the previous past pred prediction from quering the the data to more generative hypothesis from wetlab to incilico. I think again I'm very
[33:57] simplifying it. the other technologies part of the whole thing. uh I think I announced few partnership you know we do work closely with them right and see tetras is a great partner we working with them that also running on data bricks by the way right uh so the
[34:10] advantage which comes is I think you should I won't leave you a thought is a platform right um I mean if somebody tells you that hey you know we can build this layer if the platform is not sound it has no data it has no uh pipelines
[34:22] which are stable to run at a view it will it'll fall apart now I won't put any things on a agents over here because you you had a lot of download or agent you know token and cost I won't go here
[34:34] now so we kept this presentation intentionally high level please feel free to reach out to me on the agent side of it I can help you where we are in that journey right so the models uh we are building we do partner with all
[34:46] the big hyperscalers through data bricks through AI unified gateway AI gateway and then the college unified gateway so we already enabled those models so we tried to build some models or cost parameters around that how to ensure
[34:59] sure that our business folks uh the teams on the data scientists can look at those models and drive the value.
[35:11] This is a vision. Every molecule we discover, every trial we run, every patient we reach can be accelerated if you build and deploy on a unified governed AI native platform. And the good news is database platform is already enabled with open source
[35:24] frameworks. It's AI enabled. Um I mean you can use other comparable platform but this gives you a unified view how to make things happen. Okay that's pretty much you're right on time I think. 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.