Project Description

Fast Autonomous Sort,

Search of Threats, and

Exploitation of Captured Media



Because of ongoing technological advancements, the Defense Intelligence Agency (DIA) realized a need for fast autonomous sort, search of threats, and exploitation of captured media (FAS2T-RIF1). This requirement included the utilization of artificial intelligence (AI), real-time modeling, and exploitation methodologies to close the gaps between limited personnel and time constraints typically found in large- scale processing of media. In effect, the DIA needed technology outfitted with the latest machine learning (ML) algorithms designed to ingest and process large amounts of data to rapidly provide necessary information. Cape Henry Associates (CHA) utilized the Rapid Innovation Fund (RIF) Program, supported by the DIA, to take on the task of the fast autonomous sort, search of threats, and exploitation of captured media project.


DIA needed to be able to ingest and process large amounts of data and utilize the latest available ML algorithms to arrive at rapidly prototyped models that advance their objectives in data exploitation. The product had to be capable of leveraging novel AI algorithms and mathematical models to build and train against large disparate data sources as well as processing and extracting data necessary for analytical and learning activities.


RIF was key to ensuring success, as it provides a collaborative vehicle for small businesses to develop and provide innovative technology to customers. With the support of RIF, CHA collaborated with KOVA Global (KOVA) to effect the necessary technology advances to realize the goal.

Several objectives were identified and addressed:

  • Utilization of key advancements in neural networks and AI capabilities through ML, natural language processing, and deep learning;
  • Improved content delivery, including immersive, interactive model-based intelligence for what-if analysis and a focus on integration;
  • Application of discovery, analysis, and integrations against different types of captured media;
  • Delivery of platform and infrastructure solutions to ensure flexibility, scalability, and security;
  • Support of rapid prototyping to create a secure environment for the user to conduct sense-making analysis of big data, new technology insertion, small application, or scripting configuration requirements.

For product success, the above objectives were critical, as was creation of an open architecture and open application data interface (API) to ensure feature integration and data access for creating alternative solutions to large systems using modern services and capabilities available in unclassified clouds like Amazon Web Services (AWS) and Microsoft Azure. Finally, the ability to quickly spin up new instances by leveraging cloud or local environments was also identified as necessary for achieving the stated goal.

CHA’s fully managed, turnkey environment for high-volume ML computations, FogLifter, was employed to accomplish the deliverables. A stand-alone, scalable, secure, and fully dedicated mobile AI framework powered by NVIDIA NGC, FogLifter provides baseline capabilities of the current cloud ingestion and categorization pipeline with no external infrastructure requirements.


All timelines and deliverables were met. In an impressive 18 months, CHA and KOVA built a Minimally Viable Product (MVP) AI and deep learning platform that is cloud agnostic, stand-alone, and cutting-edge.

For the FAS2T-RIF1 project, FogLifter performs rapid ingestion of a wide variety of data with a user-configurable filtration of AI Interesting Artifacts (AI/IA). It allows the user free-form investigation of AI/IA as well as provides ML experimentation and training, all through a flexible visualization platform.

While this project used a limited AWS service plan processing a raw test dataset of 200GB in less than 36 hours (AWS General Purpose M5 8CPU), FogLifter has much greater data processing capability depending on several variables. For this particular instance, results can be further broken down as follows:

Image Analysis

  • Processed 36,224 images
  • Ran 181,120 jobs
  • Batches of 1,000 images
  • 5 image features
  • Results in under 2 hours using 50calls per second limit quota

Video Analysis

  • Processed 223 videos
  • Ran 1,165 jobs
  • Batches of 20 videos
  • 5 video features
  • Results in less than 24 hours using 20 concurrent jobs limit quota

Document Analysis

  • Processed 100,000+ documents
  • Batches of 10 document archives
  • 6 document features
  • Results in roughly 17 hours using 10concurrent jobs limit quota


With its FogLifter mobile AI framework powered by NVIDIA NGC, CHA produced products for this project that will expedite the DIA processing of intelligence artifacts and has effectively changed the concept of possibility for fast autonomous sort, search of threats, and exploitation in captured media. Through CHA’s collaboration with KOVA and use of the Rapid Innovation Fund, our nation’s warfighters have been provided faster access to highly relevant intelligence through skillfully created, state-of-the-art flexible technology platforms.