AI Revolutionizes State Services: Health & Traffic Transformed!

Hello, welcome everyone. Thank you for joining this lightning talk today. Um we are going to talk about uh reinventing uh state services with data bricks and specifically focus on a particular state and two um success stories one in health and the other one in transportation. I'm joined here today by Michael Powell uh who is uh section chief at California Department of Public Health. Um Rex Phillips who is a principal at our uh health and per public service strategy practice. Uh and then I am a managing director within our strategy practice for health and public service. So um let's get started with this. Um the

state that we are going to discuss today has embarked upon a fearless path of data and generative AI exploration. Uh and this has been spearheaded by the governor who released an executive order in uh September of 2023. So in partnership with the state, private sector and academia, this state is pursuing bold and complex ideas that's transforming solutions into long-lasting benefits for its residents. Um today we we are going to explore two such stories. One for the department of health at the state and the other one on the department of transportation. So um we'll start with the department of

transportation um story. Um the governor of the state recently did a press briefing uh and to quote him exactly uh generative AI has come to traffic management in the state in a way that does not exist in any other state in America. Uh this underscores the significance of the work we are doing with the department of transportation uh and essentially how this solution is being leveraged um to not just um make our roads safer but also um developing solutions that uh help with the traffic congestion within the state. Um the the other interesting point that the governor made was that uh he he perceives generative AI as an efficiency play not in form of reducing people but enabling his uh the departments within the state with AI enabled tools that allow them to serve their residents better more effectively and more efficiently. So let's let's let's talk a

little bit about the background of this department. So this is the department of transportation and they are responsible for 52,000 lane miles of state highway. They are responsible for the traffic and safety on on this highway. Now they're trying to solve for two key problems. The first one is the high congestion and the cost associated with this high congestion in the state highway systems. And this cost comes with delay in passengers and commercial vehicles uh as well as the cost which is associated with increasing emissions. So the cost is definitely twofold there. Um this cost is compounded by more than 300,000 people who are supercomputers which means they travel 90 minutes in one direction every single day. Um the second problem that this department was trying to resolve is to reduce road deaths. So on any given day in this state's transportation system, 10 people die. There are 10 to 12 road deaths on any given day. So obviously this is this is one of their critical problems and part of their vision 2050 where they're aiming for zero road deaths in the state highway system. um what they need the need of the R

hence was building a traffic mobility insights platform that provides them real time traffic data that is multimodel that is scalable and that is AI enabled and they want to leverage this traffic mobility insights platform one to analyze traffic patterns address bottlenecks and reduce one of the problems that we talked about congestion but also leverage this solution to optimize and make longlasting strategic infrastructure investments that the residents can benefit from in decades to come. So let's quickly talk about the highle solution uh for the traffic mobility insights platform. So

on the left you see the variety of data sources that this solution takes in. So that includes the uh state transportation system on traffic performance, safety um uh and and other things like transit data, but it also includes a lot of um unstructured data coming from video feeds and traffic camera feeds. Additionally, we are also ingesting thirdparty data all around transit data uh and and traffic insights. So there's a whole set of data that comes in and then there is weather data, equity data which is just again that those are like all all uh different sources. So again there's data is everything in this solution. Uh and the first two layers of our solution the infrastructure and the data layer that's where we are bringing in all these data sets together and establishing a unified governance. uh the layers on top of that the advanced analytics they cater to um uh traffic specific uh AI models that enable you to analyze patterns as well as share uh potentially counter recommendations and finally on top you have a generative AI uh interface. So there were many use cases that we built

on this. Uh most of them catered towards again traffic congestion. Uh but some of the details around um uh types of HOV lane improvements that this department could do detection of traffic uh incidents and very critical to this department emergency um evacuation management. What are the ways to manage uh evacuation uh options in case of a disaster? Um so high impact solution. Let's talk very briefly about the data. So we spoke about unified data governance in many of the sessions today. Uh this department was the opposite of it. So every every data set that you see here was a silo huge huge volume of siloed data sets. Um, PEMS as an example is their traffic performance data but that did not talk or in any way have uh any any unified uh solution with the Switter's data which is their crash data u or with any kind of transportation system network data. Add to all of this there was a need to layer in spatial analysis. So there was a need to add in geospatial data into the mix. So the insights were uh meaningful uh as well as there is an absolute need for real-time data because this is traffic management. This is real time incident management. So this makes it a very complex um data ecosystem that we had to solve for and 90% of the solution is essentially how we manage that data and we applied unified governance on that data. So the size that you see is the size that we uh uh in in form of data sets 1.5 terabytes. It was just for a proof of concept. Um so certainly there was just um uh that the actual data volume is at least like 10 to 15x that. So let's go into our solution very quickly. Um we are again no no price for guessing bringing all the data together um through a datab bricks delta lake. We are leveraging unity data catalog for bringing all um unified data governance across all the data sets that we mentioned as well as driving lineage um through them. Uh autoloader is being used to bring in data from the multiple sources. Uh and there's tremendous scale that can be accomplished from it. There's a medallion architecture established for bringing in the raw data and then processing it and g having more meaningful data sets in the semantic layer. And then finally a set of very powerful foundation models leverage this data for a variety of use cases in including um uh identifying traffic bottlenecks or proactive counter measures to traffic bottlenecks as well as proactive reduction of crashes. Um the whole road map of of the solution is to go from reactive to proactive response strategies and from manual assessment which can take days um to automated assessment through the solution that we built. So to that end the future roadmap includes instant insights that can be enabled through G gen Genie and AIBI tools all the while where a traffic engineer can converse with this solution in the natural language. Um this is a department of transportation. So they have close collaborations with universities including UC Berkeley. So data sharing options through Delta live and data marketplaces on the road map. And last but certainly not the least um for the solution to be effective the traffic operator needs realtime streaming data from the traffic cameras. So we have structured streaming in our road map to enable them to do that. This solution is

turning what was taking two to three months of work by the traffic engineers into two to three days and reducing high cost into dollars saved and most importantly crashes into lives saved. So this was the story on our department of transportation. I'll now welcome Michael to come in and talk about our department of uh health um solution. This may be the most bizarre way I've presented in a long time. Do I got good volume? Okay, awesome. So, I just want to talk a little bit about immunization

data and how with the department of public health, we were we used our data tools uh during the pandemic specifically prior to the pandemic uh immunization data although we had immunization data it wasn't required to be sent to us. Uh if immunizations were given in the state of California they usually resided with an EHR or a medical provers's office but not necessarily sent to CDPH.

During the pandemic, this all changed obviously with co vaccinations being required to be sent to the registry. We had to have a way to show this data. Uh we had a require a reporting requirement daily to the CDC for all COVID data. In addition to that, we had other tools that we needed to implement to get that data out to our local health jurisdictions and others. And as you can see illustrated on the slide where it says immunization data. Now, when the pandemic started, that was our immunization data. That's all we had. Since then, we've implemented the digital vaccine record. If you're a resident of the state of California, hopefully you've used this tool to get your own immunization record. If you haven't, hopefully your state has that as well. Uh, with that digital vaccine record, we also have other tools for the virtual assistant for people who couldn't find their vaccination record or needed assistance in getting that record. Among all of this, there were a lot of other things that we wanted to do with the data that we needed to do with the data and we didn't have the tools to do that. And that's where we partnered with Accenture to help build those tools. And with that, I'm going to turn it over to Rex to go into more of the technical solution. He's going to be much more interesting than me anyway. All right. So the key point of what

we're going to go for here is to build a solution that would leverage data bricks to do some very important processing which is uh powered by data bicks's ML capability using MLflow and it's to uh expose this now as a software as a service. We called it analytics as a service. uh key to doing this was to first process everything that we had in data bricks and then to expose it in using our cloud-based solution with an API. So in this case we had uh databicks do the processing exposing it to a NoSQL graph database which would then be shared with um in this case we use Azure Azure API gateway. Also, while we did this, we wanted to make sure we could expose this to any potential okay, expose us to any potential um clients or other markets that we have. So there we leverage the delta sharing with the hope of then applying the marketplace uh going forward. Okay. So the actual layout of this so what you see here is a multiple of different inputs that we had coming in from the left. We did use Azure data factory to consume things. Now the reason we use Azure data factory uh in this case instead of some of the native capabilities is we had to do orchestration across multiple different systems. So Azure data factory is handling our orchestration. It's also ingesting the data. It's loading it into our ADLS uh where we're then ingesting it using autoloader uh putting it through medallion architecture and this is also where our ML processing happens. You know what's unique about this solution

is we're doing patient matching or the master person index. Master person index is essentially a 50 million person categorization task. We have 50 million people in our system. We're trying to match those together in order to make sure that they're all uniquely identified. So we go through the process. We load our data in. We run it through medallion architecture. We then do our uh ML powered matching. Uh after we get that, that's when we're then going to load it back in this case to ADLS where it can then be consumed by all of our different groups. So this is the data that goes to the CDC. It goes to our API gateway which are then consumed by some of the other different um groups that we have available. By doing it this way, we're able to leverage the power of data bricks to power one of the most powerful masterperson indexes available in any state. Um, which is truly a marvel. So, to kind of I'm going to turn this over to Michael again to kind of wrap this up. Using the power of data bricks, we have been able to do a massive system uh to handle all the different members of this state, to handle all the different immunizations uh for this state, to handle 50 million people. Uh and it's truly been a work and a wonder. Thanks, Rex. So, I'm going to close it

out just by giving some numbers of what we were able to accomplish with this solution. Um, as Rex mentioned, we have over 50 million patients in the immunization registry. I know that's greater than the population of California, but it is a life-to- registry. And people who move in and out of the state, they'll they'll remain in our registry forever. Uh, we have over 20 million vaccination records that were downloaded by individuals throughout the state. We have eight language translations for the digital vaccine record. Uh, and the digital vaccine record solution that we implemented, six other states have adopted that same solution. although either in part or maybe slightly modified but they are using the solution that we developed the code for and another uh big accomplishment was uh 1 hour from vaccine uh is submission to reporting um with the number of records we have um 680 million doses in our registry that's a really big task to be able to report that quickly so all in all I can say that this was a successful uh endeavor um we've had a lot of success we hope to continue the success and we won't talk about funding in this meeting but you know u we hope to continue that success and everything that we've learned over the last 5 years and with that I'll close. Thank you very much for your time and attention today.