SOAR is the IOS behavioral health decision support platform, built for military, veteran, and civilian healthcare organizations. The platform supports risk identification, monitoring, and clinical documentation for care teams. Development is supported through a DARPA Phase II program, with clinical subject matter expertise provided by Columbia University.

A clinician-facing conversational interface, speech and language analysis, structured clinical documentation, and longitudinal patient history come together in a single working environment. Clinicians see the evidence behind every indicator, including the specific language, response patterns, and history that contributed to a flag. Behavioral health professionals need transparency into how a recommendation was generated, both to apply their own judgment and to produce a defensible record of how a determination was reached. SOAR presents the contributing factors rather than a score alone.

SOAR was designed with military and veteran populations in mind, where deployment cycles, stigma, and fragmented records complicate risk detection. The same architecture applies to VA and DHA care settings, state behavioral health programs, operational medicine, correctional and campus health, crisis response centers, and broader digital health modernization efforts.

Privacy is handled through federated learning and de-identification, allowing models to improve across sites without moving protected health information between them. The platform runs on a FedRAMP-aligned cloud architecture with an ATO-ready deployment pipeline.

Interoperability is handled through standards-based interfaces. SOAR is designed to exchange data with HL7 and FHIR enabled clinical systems and to operate alongside an existing electronic health record rather than replace any part of it.

• Behavioral health risk identification and triage support
• Suicide risk assessment aligned to established clinical instruments, including the C-SSRS
• Clinician-facing conversational interface for intake and structured assessment
• Clinician co-pilot for documentation and case review
• Speech and natural language analysis
• Longitudinal patient monitoring and change detection
• Explainable AI with every indicator traceable to its source evidence
• Federated learning and de-identification for multi-site deployment
• Standards-based interfaces for HL7 and FHIR enabled clinical systems
• Electronic health record integration
• Integration with existing clinical documentation and care workflows

IOS decision support systems fuse multiple data sources into a single assessment and present the reasoning behind it. Sensor feeds, imagery, telemetry, text reporting, and audio are combined into a common picture, then reasoning models surface what changed, what it likely means, and how much confidence the assessment carries.

Uncertainty is shown rather than hidden. Where a recommendation depends on a weak, stale, or contradicted input, the operator sees that dependency. IOS work in this area includes neurosymbolic approaches that pair learned perception with formal logic, producing machine-checkable reasoning traces instead of unexplained outputs. That matters wherever a decision will be reviewed after the fact or an automated recommendation has to be defended.

• Multimodal data fusion across sensor, imagery, text, and audio sources
• Risk assessment and course of action analysis
• Mission planning and rehearsal support
• Situational awareness and change detection
• Explainable and auditable reasoning traces
• Confidence and uncertainty reporting at the point of decision
• Integration with existing command and control and analytic environments

Sensing programs more often stall on data handling than on sensor performance. IOS develops the processing layer that converts raw sensor output into information an analyst or an autonomous system can use.

Work spans electro-optical and infrared imagery, RF signal analysis, and multi-sensor fusion, with computer vision models for detection, classification, and tracking. Processing is designed to run where the data is generated, including platforms with limited power, thermal budget, and network access. IOS is an Altera partner with published design offerings in sensor fusion and edge AI, which gives programs a defined path from algorithm development to accelerated implementation on fielded hardware.

• Electro-optical and infrared image processing
• RF signal analysis and characterization
• Computer vision for detection, classification, and tracking
• Multi-sensor fusion and cross-sensor correlation
• Perception software for autonomous and uncrewed platforms
• Onboard processing for size, weight, and power constrained systems
• Sensor data pipelines, labeling, and model evaluation workflows

Model performance in a data center says very little about performance on a platform. IOS engineers AI systems for the hardware the program actually has and the connectivity the mission actually provides.

Capability covers FPGA design and hardware acceleration, GPU inference optimization, embedded deployment, and model compression for constrained targets. Systems are built to keep working through degraded or denied connectivity, holding local inference and reconciling with enterprise systems when a link returns. IOS maintains published FPGA design services, AI acceleration, and edge AI offerings through the Altera partner program and has applied these approaches under Department of Defense research programs.

• FPGA design, integration, and hardware acceleration
• GPU and accelerator inference optimization
• Embedded and tactical edge deployment
• Model compression, quantization, and pruning
• Operation through disconnected and intermittent connectivity
• Hardware-in-the-loop testing and validation
• Cloud to edge model management and update pipelines

Programs increasingly require engineering artifacts to be models rather than documents. IOS supports that transition with toolchain integration and automation that connect authoritative models to the systems and reviews that consume them.

Support includes MBSE authoring and tool integration, digital twin development for simulation and analysis, requirements traceability from need through verification, and automated generation of engineering artifacts from the model of record. Emphasis stays on keeping the model and the delivered system in agreement over time, which is where most digital engineering efforts lose their value.

• MBSE toolchain setup, integration, and automation
• Digital twin and simulation environment development
• Requirements management and end to end traceability
• Interface control and architecture modeling
• Automated generation of engineering artifacts and documentation
• Model to code and model to test workflows

Most of the value an agency fails to get from AI is blocked by integration rather than by model quality. IOS builds and integrates the software around the model so that a working prototype becomes a system people can actually use.

Teams deliver cloud-native and hybrid applications, secure DevSecOps pipelines, API and data layer development, and integration with legacy government systems. IOS is CMMC Level 2 self-certified and builds to open standards and documented interfaces, so a customer keeps the ability to change direction or change vendors without discarding the work.

Capabilities

• Custom application development, cloud-native and hybrid
• DevSecOps pipelines with automated security scanning and testing
• Kubernetes and container platform engineering
• API design, data engineering, and system interfaces
• Legacy system integration and modernization
• Enterprise content management and search modernization
• Identity, access, and secure deployment engineering

Emerging capability is only useful if it survives the move out of the laboratory. IOS operates as a research performer and as a transition partner, holding awards and agreements with DARPA, the Air Force, the Space Force, and other research sponsors, and carrying that work forward into deployable software.

Active research spans behavioral health AI, neurosymbolic reasoning, biological computing, and air and space domain applications. Prototype work uses the same engineering practices as production delivery, so a successful demonstration does not have to be rewritten before it can be fielded.

• SBIR, STTR, BAA, and Other Transaction research execution
• Rapid prototyping and technology demonstration
• Technology transition and maturation planning
• Teaming with universities, national laboratories, and primes
• Independent research and development in AI and intelligent sensing