SBIR-STTR Award

ATA’s proposed innovation, SOAP, is a platform for developing, deploying, monitoring, and maintaining machine learning-integrated streaming data pipelines. The solution will simplify the historically time-consuming, manual, and code heavy tasks associated with operational use of machine learning (ML) and streaming data management, including integrating new data sources, creating data pipelines, and applying ML. Leveraging a portable infrastructure and distributed processing and storage, SOAP offers the scalability needed to operate on the large volumes, velocities, and varieties of data typical of modern streaming data applications, including multi-sensor integration, digital engineering, and model and simulation. SOAP addresses the relative lack of investment into operationalizing streaming ML by directly providing a means for data engineers and scientists to create, monitor, and update ML-integrated data pipelines. SOAP includes two primary innovations: 1) services to execute ML models within lightweight, easily configurable streaming data pipelines, and 2) a service for monitoring deployed ML model performance for degrading accuracy, stability, and speed. We propose to develop the new services on top of mature, widely-adopted stream processing technologies while extending the functionality of more basic Machine Learning Operations (MLOps) tools. By promoting extensibility and ease of integration with outside data and services, this design is cost effective, leverages prior investment, and avoids the technical debt associated with single-tool data soloing. Approved for Public Release | 21-MDA-11013 (19 Nov 21)

ATA’s proposed innovation is SPOT-lyt, the Streaming Platform for Object Tracking Analytics, a system for real-time and forensic tracking and analysis of moving objects with cutting-edge capabilities in data fusion, simulation, machine learning, and streaming analytics platform operations. For Phase II of this SBIR effort, we propose to use the prototype developed during Phase I as the foundation for this innovation, addressing current MDA needs in object monitoring, surveillance, and intelligence while also incrementally progressing the Department of Defense's experiments in applying machine learning to multi-sensor integration, digital engineering, model and simulation, and big data. During Phase I, ATA tested the feasibility of a platform for developing, deploying, monitoring, and maintaining ML-integrated streaming data pipelines for use in real-time object tracking and analysis. The prototype demonstrated the feasibility of a system that simplifies the historically time-consuming, manual, and code-heavy tasks associated with operational use of ML and streaming data management with innovations that 1) support the ingestion of multiple streaming data sources into the platform using a flexible model to quickly onboard additional new data sources, 2) utilize low-code configuration of ML-integrated streaming data pipelines for simpler streaming data pipeline development and deployment, 3) leverage automation and other operations-related services to support the development, testing, and deployment of ML models, 4) monitor deployed ML models to account for degrading model performance, data drift, and unexpected data, and 5) generate synthetic data to simulate detection events and publish the event data into the system for platform testing and training. The overall goal of Phase II is to mature the Phase I prototype (Technology Readiness Level, TRL 4) into a robust, detailed SPOT-lyt prototype of TRL 6 (Full System Prototype in a Relevant Environment). By the end of the phase, SPOT-lyt will have the additional capabilities required of an operational object tracking system and the platform portability, security, and reliability needed for adoption into production environments. Approved for Public Release | 22-MDA-11340 (16 Dec 22)