Every day, satellites, sensors and other devices collect hundreds of terabytes of imagery data. These images hold incredibly powerful information. Mining them for intelligence to protect American interests at home and abroad is critical to national security.
As a long-time partner to government agencies, GDIT supports the enhancement of the analytical capabilities used to process huge and ever-growing volumes of imagery. Much of that work involves ensuring analysts can leverage the latest in Artificial Intelligence (AI), increasingly known as Super Intelligence (SI), to accurately and comprehensively analyze complex and diverse image sets.
Our team was asked to make the most efficient use of an intelligence customer’s AI resources — from the environments necessary to develop and pilot AI-based solutions, to managing their cloud spend and ensuring compute resources were leveraged to the fullest. The customer recognized that both its AI use and its resources were compartmentalized and siloed, and that a unified enterprise approach would not only ensure more efficient AI access and availability but would also enhance its ability to deliver on its critical national security mission.
From the start, we knew there were two critical areas where we could help. First, we provided the customer with our Machine Learning Operations (MLOps) platform to help them build custom AI/ML workflows. Today, we maintain the platform and keep it accessible and secure, ensuring the customer has access to the latest, proven and secure AI/ML tools. By applying proven DevOps methods, the MLOps platform speeds innovation, reduces time to deployment and drives accuracy and efficiency of AI/ML solutions – which immediately improves the customer’s ability to rapidly build and deploy them.
Second, we offered the customer access to our High-Performance Computing (HPC) as a service offering. Internally called Enterprise HPC, this environment gives users a place to house their high-performance compute needs, essential for tasks like image analysis and computer vision. Analysts can use inference services and large language models to mine images and create reports that support critical decision making. Leveraging an integrated suite of open-source and commercial tools, Enterprise HPC supports the analyst and their multimodal workflows.
By making these capabilities available to the customer, including our HPC Integration and Performance Engineering (HIPE) service, we expanded their use of AI while streamlining AI/ML projects. HIPE bridges the gap between users and Enterprise HPC resources, optimizing workflows for the customer’s specific HPC environment. For example, HIPE cut inference time in one workflow from 17 days to 17 minutes and saved an estimated $3 million to $4 million annually. Another workflow migration reduced delivery time from two days to under two hours, with an estimated $1.5 million in savings annually.
The more the customer uses AI within its workflows, the better the models perform and the easier it becomes to integrate new solutions. This means that sharper insights from large, complex datasets reach analysts faster to support data-informed decisions.
These gains extend across our portfolio. Our Dynamic Resource Integration for Versatile Execution (DRIVE) program further amplified computing power by pooling thousands of idle GPUs into a distributed network, enabling 24/7 utilization at minimal cost. Combined with AI-driven code generation, these efforts deliver enterprise-level scalability—turning desktop models into optimized, high-performance solutions. By leveraging idle hardware, we created a massive increase in computing power.
We take the same approach in our own operation, using AI for code generation and model augmentation to accelerate bespoke application development with highly specific needs in the intelligence community. This drives enterprise-level model scalability: data scientists build models on their desktop, and we optimize and scale them to the enterprise.