Sponsored by: Accenture | Securing America’s Scientific Future

Summary

Accenture is building a Unified Discovery Platform for the Department of Energy's Genesis Mission, connecting over 40,000 scientists and engineers across 17 national labs and 700+ partner organizations with the stated goal of doubling the productivity of American science within a decade.
The platform's medallion architecture runs from a bronze immutable data store through to gold AI-ready data, with a purpose-built Agentic Research Assistant featuring 8 tasks, 206 subtools, and access to 198 data sources built on Mosaic AI, Unity Catalog, and Lakeflow.
This video describes a tiered identity model supporting hundreds of collaborating organizations and mission apps that go from requirements to prototype in one week, spanning a full architecture that includes supercomputing, quantum, and autonomous lab integrations.

Chapters

FAQs

What is the Department of Energy's Genesis Mission?

The Genesis Mission is a DOE initiative that brings together over 40,000 scientists and engineers from 17 national labs, along with more than 700 academia and private industry partners, to accelerate scientific discovery using AI at scale. This video describes it as the first time America's scientific community has unified around a shared platform since the space race, with a KPI of doubling scientific productivity within a decade.

How does the Unified Discovery Platform support so many collaborating organizations securely?

The platform uses a tiered identity model designed to manage access for the 700+ organizations participating in the Genesis Mission. This video explains that the architecture accommodates the varying trust levels and data-sharing agreements among national labs, academic institutions, and private industry partners while maintaining security and auditability through Unity Catalog governance.

What is the Agentic Research Assistant built for the Genesis Mission?

The Agentic Research Assistant is a purpose-built AI agent described in this video as supporting 8 research task types, 206 subtools, and access to 198 data sources. Built on Mosaic AI, it is designed to help scientists navigate complex research data, run analyses, and surface insights across the unified data platform without requiring expertise in every connected system or dataset.

What is the scale goal for the Genesis Mission's impact on American science?

The stated KPI for the Genesis Mission, as described in this video, is to double the productivity of American science within a decade. This goal is backed by 26 science and technology challenge areas identified by the DOE as pilot lanes for AI implementation, encompassing projects across energy, materials science, and other strategic research domains.

Full transcript

[00:08] My name is Wyatt Sullivan. I'm an enterprise architect with the Center for Federal Services. And today I'm going to be talking about securing America's scientific future. Over the last couple months, I've been supporting the Department of Energy's Genesis mission. Before I get started though and give you
[00:21] a little bit of context about the mission, we're going to kick things off with a brief video.
[00:45] >> Discovery always drives progress. >> We choose to go to the moon in this decade and do the other things >> not because they are easy but because they are hard. Today, the pace of discovery can be daunting.
[00:59] Sure, data is everywhere, but it lives in silos. When systems don't connect, breakthroughs take time. And time waits for no one.
[01:16] But now there's a new way to unlock innovation. One that binds AI to advanced computing and secure infrastructure, not as separate data sets, but as one unified platform.
[01:32] Accenture is
[01:46] scalable environments. Where researchers spend less time navigating complexity and more
[02:00] America's scientific future. Forged in the labs, proven across the mission, already answering the science. We're expanding collaboration, scaling innovation from idea to impact, building
[02:13] the foundation for the
[02:30] Now, you may be wondering about a little more context about all of that. It
[02:43] scientific organizations from across the federal government. That totals over 40,000 scientists and engineers, best-in-class instruments and data, decades and decades of data, all coming
[02:55] together to align on one mission, accelerating scientific discovery with AI. This is the first time that America has come together like this to accelerate a process. The last time the American
[03:08] scientific engine came together like this, it was the space race. And the time before that was the Manhattan project. When naming the Genesis mission, they called back to big projects like that. But they decided to go with something new because this is AI
[03:21] at scale like never seen before. The Department of Energy kicked things off by identifying 26 science and technology challenges that are pilot lanes for different projects that
[03:34] encompasses countless projects and countless experiments that need to be driven forward by the implementation of AI. Oh, and did I mention there's over 700 academia and private industry partners all coming to the table to help
[03:47] drive this forward? The KPI is simple. double the productivity of American science within a decade. That's a call to action like we've never seen before. And this is a government
[03:59] mission like we've never seen before. So we at AFS asked ourselves three questions when we decided how to help out. First, the scientific wealth of America is scattered across multiple
[04:12] different places across governments, across the GA uh labs, across academia, and across private industry. What could researchers achieve if we brought that all into one secure ecosystem and made it easy to collaborate on? Next,
[04:26] state-of-the-art tools aren't where they're needed to be. Scientists right now have to act as their own infrastructure and platform teams to get the tools they need in place to accelerate scientific outcomes. What could they achieve if they didn't have
[04:38] to spend all that time in implementation? And third, applying science is really hard. How do we get the model out of the notebook? How do we get the insights out of the paper? How do we apply the
[04:50] science to real world use cases and provide value to the American citizens? And last is a realization in the scientific process. 99 failures and 99 different tries. All this different data
[05:02] is all coming into one place, but it's left on the cutting room floor when you publish a paper. That one really great data science breakthrough or model gets published and the rest of that is left and doesn't go to use for the American
[05:15] scientific community. So our solution is coming up with a platform that we've decided to name the unified discovery platform or UDP for short. Now as a technologist it was a little
[05:27] bit of a humbling experience when we started to design this out because we realized that we didn't really need to build anything new. The best-in-class platforms and tools already existed across industry. We put four different
[05:39] layers together, an easy to use user interface that executive stakeholders can come in and see insights, but researchers can get hands-on and get dirty with the data and insights and things that they need to do within the
[05:51] platform. Additionally, we have scientific modeling and workflow automation that makes all of this scalable without writing infrastructure as code. We have AI or agent orchestration through Mosaic AI that
[06:03] powers and governs all of the agents within our platform. And all of this is built on a strong data backbone and data integration and governance layer that standardizes and secures everything with the unity catalog and lakeflow
[06:15] pipelines. In addition to data bricks, we also have other partners coming to the platform. AWS is our infrastructure partner. We have frontier models provided by OpenAI capability multipliers like Nvidia, AMD
[06:28] and other partners from across technology uh the technology landscape. And then we've also planned the platform to scale to a multicloud architecture with advanced security features as well. A single integrator and a unified
[06:40] experience for all the scientists across the United States. Now what's the first problem you need to solve in any platform? Identity. And this isn't just a normal identity
[06:52] problem. This is 700 collaborators and organizations all coming together in one place and needing to collaborate on secure data. We developed a tiered identity structure that brings together
[07:04] the government folks, people from academia, private partners, and verifies their identity and pulls them all in through a tiered identity structure that uses role-based and attribute-based controls that propagate from the login
[07:16] screen all the way down through your tools, data, and infrastructure. This is one way to access everything. Now, putting my researcher hat on, I'll log in.
[07:32] This is your landing page. It looks simple. It is simple. And that's precisely the point. Scientists can get to the science faster and we handle the technology and governance on the back end. Clicking into a project workspace, you'll see a team's home base here. So,
[07:44] this is where everything comes together and people can collaborate. Kicking off a project could be hard. The baseline knowledge exists, the data exist, the agents and tools and research workflows may exist as well. But how do you get to
[07:57] it? And how do you pull it together all in one place? Our platform brings together open- source data, enterprise data, agentic workflows, traditional ML tooling, collaboration tools, all in one
[08:11] place with access to the Department of Energy's supercomputing capabilities, cloud-based superco computing capabilities, GPU accelerators, quantum capabilities, all powering forward American science with AI.
[08:28] Taking a look at the data layer that makes this all possible. We have a tiered medallion architecture, bronze, silver, and gold. We like to keep it simple so it's easy to digest. Bronze is an immutable data storage that when you upload things, it's a scientific store
[08:41] that doesn't change. It doesn't um it doesn't get an analysis or transformations applied to it. Silver is that project aware data set. This is where those hard to solve problems are like and iteration cycles go uh through
[08:55] the data. And then gold is decision ready AI ready data sources that can be published to the enterprise or out to public um public repos. So we've designed a data and modelmart to make it uh possible to share with the enterprise
[09:07] and we've enabled delta sharing to share it out to other government organizations to make that data data accessible for their use cases and provide value to the American citizens. All this is made possible through a Gentic ETL that the
[09:20] agent drafts in a human approves for those pipelines and it all feeds through in a streamlined way and is enforced um through the Unity catalog and policies. This is a view of that data mart. We
[09:33] have different data cards there that makes it easy to digest the data that's in the platform. It gives a brief description, the data type and a little bit of information about the data. It's easy to slice through this through the enter the the data mart search tool as
[09:46] well as filtering and then it's easy to add to the project. Once you add it to the project, it then becomes a silver data set and you can apply analysis and transformation without messing with the data that's in the gold data.
[09:58] Models get the same treatment in its sister model and that all propagates throughout the platform. Now taking a look at some of our agentic tooling. This is our flagship Agentic Workflow, which is the Agentic Research
[10:11] Assistant. And it may look like something you've seen before, but it offers additional capabilities. On the left, you'll see it's selecting the different data sources that it's pulling into the platform. And this will only grow as we scale this solution. We're
[10:23] going to ask it a question from our critical minerals use case that we talked a little bit about before. And it's about X-ray light sources and material science and considering techniques that have been abandoned in the past. The first thing the agent does
[10:35] is consider a plan and walk you through its steps that it's planning to execute. And you'll see what it's thinking here in a moment. So there's the seven steps it's planning to execute. And then you can approve and review what it's looking to do. From there, it'll begin to
[10:49] execute. And this keeps the human in the lead element of the research workflow. It keeps the expert in the driver's seat, but it allows them to get to the answer faster. You'll see it's running through
[11:01] its tasks. It's running through the tools it's using. And then on the far far right there, you'll see the citations that it's pulling in as well. All of this comes together in one report that allows the uh researchers to do
[11:13] deep analysis of those hard to solve questions. The numbers behind this uh once it's run all the way through, you'll see it ran eight tasks. It utilized 206 subtools and agents. It referenced 198 different data sources
[11:26] and pulled in the most relevant 14 of those. That's the difference between a regular workflow or an off off-the-shelf tool and a scientific workflow. In addition to this tool, we also have a couple other tools available in the
[11:38] platform today, which is an uh EDA agent or exploratory data agent that allows easy exploration of the data and transformation. And then we also have a report generation tool that removes the administrative burden from researchers.
[11:51] We have planned integrations with multiple DOE agentic workflows as well as private industry uh partner offerings that we plan to onboard into the platform. This will just compound the effect and scale of the data as we bring
[12:04] more and more resources on and these agents learn more and more about the data that's underlying the platform. Now we realize something and as we have 26 different uh SMT challenges that we
[12:16] have feeding into this platform, UDP is not a solution for all of them. We recognized the need for custom solutions across all these different challenges and we've developed something called mission apps. This is a mission app
[12:28] that's also based on our critical minerals use case and we developed this in partnership with the critical minerals team. We met with them on a Wednesday and by the next Wednesday we had a prototype ready to show them. They just gave us some requirements. We used
[12:40] the tools that are natively available in the platform and we had a first draft solution ready to show them the next week. that shortens the time from prototype to innovation dramatically. In American science, there's multiple
[12:52] different ways we can implement things like this. One of them is the Genie tool. That's kind of the tier one. And then we have custom containerbased applications that can be deployed as well.
[13:04] Taking a look at the architecture that makes this all possible. This is a really highle view of the platform. On the left, you'll see the different sources that we have feeding in. We have the DOE facilities and instruments. We have federal open source data stores like edx and data.gov. And then we have
[13:17] lab and partner repositories all feeding in from partners like AWS open data and data bricks delta sharing all feeding into the platform quickly and easily. In the center you'll see kind of the trust fabric of the whole platform with
[13:29] lakeflow pipelines running the ingestion that feeds into that medallion architecture that I talked a little bit about already. And then we have the unity catalog is the catalogic of record enforcing those identity policies throughout the entire platform. Our
[13:41] governent workflows use mosaic AI and then we're serving all of this up with Genie SQL APIs and lakebased sync tables. On the right you'll see some of the outcomes of this but I want to highlight two big things there which is that advanced compute that I mentioned
[13:54] all those supercomputers quantum computers all available to those 40,000 plus users on an accessaware and projectaware basis. This is all one job orchestration tool all feeding throughout the platform. And then on the
[14:06] right there we have something called autonomous labs which we partner with our emerald cloud labs to develop. This is essentially uh autonomized robotic lab capabilities. You submit your job in the cloud or submit your experiment. The
[14:19] robot robotic lab runs it and returns the data for output. This keeps the experimentter in a position where they can just keep designing new experiments and running them and getting results much faster than traditionally. You don't have to request access to an
[14:31] instrument in a different lab. you don't have to um put a request form or reach out to somebody to see if you can use something. You can just submit that job and get access almost in real time. So what does all of this mean and what is
[14:43] the platform outcomes? We have a couple different things here. For scientists, it means that that hypothesis to prototype to decision-m goes from years to weeks. That the path to a prototype and AIdriven tooling is
[14:57] dramatically improved. for the middle there for academia. We've allowed the all of this data to flow into the platform in an IPaware way. So you get credit where credit is due and you can
[15:09] see the contributions you've made in a project. And then for industry, this serves as a bridge for the gap to fast and effective public and private collaboration. Essentially within the platform, we've identified that tiered identity model
[15:22] and have planned collaborations with other government organizations to make this an identity model that can scale across the government. I'll leave you on a few things here. I want to say a big thank you to our partners at Data Bricks for the uh
[15:34] platforms and tools that made this all possible. And rolling on the momentum of this platform, we have a data bricks at the forge event on July 15. Please feel free to scan that QR code if you'd like to attend and we can reach out and
[15:46] collaborate with you. And then from there, if you'd like more information on what AFS is doing for the Genesis mission, please scan the second uh QR code there. I invite you to join us in accelerating scientific discovery and helping to translate this all of this
[15:59] scientific value into value for American citizens. Thank you very much and I look forward to meeting with you at the Asenture booth potentially after this session. Thanks.

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.