This blog will explore how the joint solution from DataRobot and Deepwave — powered by NVIDIA — delivers a secure, high-performance AI stack, purpose-built for air-gapped, on-premises and high-security deployments. This solution ensures agencies can achieve genuine data sovereignty and operational excellence.
The need for autonomous intelligence
AI is evolving rapidly, transforming from simple tools into autonomous agents that can reason, plan, and act. This shift is critical for high-stakes, mission-critical applications such as signals intelligence (SIGINT), where vast RF data streams demand real-time analysis.
Deploying these advanced agents for public and government programs requires a new level of security, speed, and accuracy that traditional RF analysis solutions cannot provide.
Program leaders often find themselves choosing between underperforming, complex solutions that generate technical debt or a single-vendor lock-in. The pressure to deliver next-generation RF intelligence does not subside, leaving operations leaders under pressure to deploy with few options.
The challenge of radio intelligence
Signals intelligence, the real-time collection and analysis of radio frequency (RF) signals, spans both communications (COMINT) and emissions from electronic systems (ELINT). In practice, this often means extracting the content of RF signals — audio, video, or data streams — a process that presents significant challenges for federal agencies.
- Modern RF signals are highly dynamic and require equally nimble analysis capabilities to keep up.
- Operations often take place at the edge in contested environments, where manual analysis is too slow and not scalable.
- High data rates and signal complexity make RF data extraordinarily difficult to use, and dynamically changing signals require an analysis platform that can adapt in real-time.
The mission-critical need is for an automated and highly reconfigurable solution that can quickly extract actionable intelligence from these vast amounts of data, ensuring timely, potentially life-saving decision-making and reasoning.
Introducing the Radio Intelligence Agent
To meet this critical need, the Radio Intelligence Agent (RIA) was engineered as an autonomous, proactive intelligence system that transforms raw RF signals into a constantly evolving, context-driven resource. The solution is designed to serve as a smart team member, providing new insights and recommendations that are far beyond search engine capabilities.
What truly sets the RIA apart from current technology is its integrated reasoning capability. Powered by NVIDIA Nemotron reasoning models, the system is capable of synthesizing patterns, flagging anomalies, and recommending actionable responses, effectively bridging the gap between mere information retrieval and operational intelligence.
Developed jointly by DataRobot and Deepwave, and powered by NVIDIA, this AI solution transforms raw RF signals into conversational intelligence, with its entire lifecycle orchestrated by the trusted, integrated control plane of the DataRobot Agent Workforce Platform.
Federal use cases and deployment
The Radio Intelligence Agent is engineered specifically for the stringent demands of federal operations, with every component built for security, compliance, and deployment flexibility.
The power of the RIA solution lies in performing a significant amount of processing at the edge within Deepwave’s AirStack Edge ecosystem. This architecture ensures high-performance processing while maintaining essential security and regulatory compliance.
The Radio Intelligence Agent solution moves operations teams from simple data collection and analysis to proactive, context-aware intelligence, enabling event prevention instead of event management. This is a step change in public safety capabilities.
- Event response optimization: The solution goes beyond simple alerts by acting as a digital advisor during unfolding situations. It analyzes incoming data in real-time, identifies relevant entities and locations, and recommends next-best actions to reduce response time and improve outcomes.
- Operational awareness: The solution enhances visibility across multiple data streams, including audio and video feeds, as well as sensor inputs, to create a unified view of activity in real-time. This broad monitoring capability reduces cognitive burden and helps teams focus on strategic decision-making rather than manual data analysis.
- Other applications: RIA’s core capabilities are applicable for scenarios requiring fast, secure, and accurate analysis of massive data streams – including public safety, first responders, and other functions.
This solution is also portable, supporting local development and testing, with the ability to transition seamlessly into private cloud or FedRAMP-authorized DataRobot-hosted environments for secure production in federal missions.
A deeper dive into the Radio Intelligence Agent
Imagine receiving complex RF signals analysis that are trusted, high-fidelity, and actionable in seconds, simply by asking a question.
DataRobot, Deepwave, and NVIDIA teamed up to make this a reality.
First, Deepwave’s AIR-T edge sensors receive and digitize the RF signals using AirStack software, powered by embedded NVIDIA GPUs.
Then, the newest AirStack component, AirStack Edge, introduces a secure API with FIPS-grade encryption, enabling the deployment of signal processing applications and NVIDIA Riva Speech and Translation AI models directly on AIR-T devices.
This end-to-end process runs securely and in real-time, delivering extracted data content into the agent-based workflows orchestrated by DataRobot.
The solution’s agentic capability is rooted in a sophisticated, two-part system that leverages NVIDIA Llama-3_1-Nemotron-Ultra-253B-v1 to interpret context and generate sophisticated responses.
- Query Interpreter: This component is responsible for understanding the user’s initial intent, translating the natural language question into a defined information need.
- Information Retriever: This agent executes the necessary searches, retrieves relevant transcript chunks, and synthesizes the final, cohesive answer by connecting diverse data points and applying reasoning to the retrieved text.
This functionality is delivered through the NVIDIA Streaming Data to RAG solution, which enables real-time ingestion and processing of live RF data streams using GPU-accelerated pipelines.
By leveraging NVIDIA’s optimized vector search and context synthesis, the system allows for fast, secure, and context-driven retrieval and reasoning over radio-transcribed data while ensuring both operational speed and regulatory compliance.
The agent first consults a vector database, which stores semantic embeddings of transcribed audio and sensor metadata, to find the most relevant information before generating a coherent response. The sensor metadata is customizable and contains critical information about signals, including frequency, location, and reception time of the data.
The solution is equipped with several specialized tools that enable this advanced workflow:
- RF orchestration: The solution can utilize Deepwave’s AirStack Edge orchestration layer to actively recollect new RF intelligence by running new models, recording signals, or broadcasting signals.
- Search tools: It performs sub-second semantic searches across massive volumes of transcript data.
- Time parsing tools: Converts human-friendly temporal expressions (e.g., “3 weeks ago”) into precise, searchable timestamps, leveraging the sub-10 nanosecond accuracy published in the metadata.
- Audit trail: The system maintains a complete audit trail of all queries, tool usage, and data sources, ensuring full traceability and accountability.
NVIDIA Streaming Data to RAG Blueprint example enables the workflow to move from simple data lookup to autonomous, proactive intelligence. The GPU-accelerated software-defined radio (SDR) pipeline continuously captures, transcribes, and indexes RF signals in real-time, unlocking continuous situational awareness.
DataRobot Agent Workforce Platform: The integrated control plane
The DataRobot Agent Workforce Platform, co-developed with NVIDIA, serves as the agentic pipeline and orchestration layer, the control plane that orchestrates the entire lifecycle. This ensures agencies maintain full visibility and control over every layer of the stack and enforce compliance automatically.
Key functions of the platform include:
- End-to-end control: Automates the entire AI lifecycle, from development and deployment to monitoring and governance, allowing agencies to field new capabilities faster and more reliably.
- Design Architecture: Purpose-built with the NVIDIA Enterprise AI Factory architecture, ensuring the entire stack is validated and production-ready from day one.
- Data sovereignty: DataRobot’s solution is purpose-built for high-security environments, deploying directly into the agency’s air-gapped or on-premises infrastructure. All processing occurs within the security perimeter, ensuring complete data sovereignty and guaranteeing the agency retains sole control and ownership of its data and operations.
Crucially, this provides operational autonomy (or sovereignty) over the entire AI stack, as it requires no external providers for the operational hardware or models. This ensures the full AI capability remains within the agency’s controlled domain, free from external dependencies or third-party access.
Specialized collaborations
The solution is a collaboration built on a co-developed and enterprise-grade architecture.
Deepwave: RF AI at the edge
DataRobot integrates with highly skilled, specialized partners like Deepwave, who provide the critical AI edge processing to convert raw RF signal content into RF intelligence and securely share it with DataRobot’s data pipelines. The Deepwave platform extends this solution’s capabilities by enabling the next steps in RF intelligence gathering through the orchestration and automation of RF AI edge tasks.
- Edge AI processing: The agent uses Deepwave’s high-performance edge computing and AI models to intercept and process RF signals.
- Reduced infrastructure: Instead of backhauling raw RF data, the solution runs AI models at the edge to extract only the critical information. This reduces network backhaul needs by a factor of 10 million — from 4 Gbps down to just 150 bps per channel — dramatically improving mobility and simplifying the required edge infrastructure.
- Security: Deepwave’s AirStack Edge leverages the latest FIPS mode encryption to report this data to the DataRobot Agent Workforce Platform securely.
- Orchestration: Deepwave’s AirStack Edge software orchestrates and automates networks of RF AI edge devices. This enables low-latency responses to RF scenarios, such as detecting and jamming unwanted signals.
NVIDIA: Foundational trust and performance
NVIDIA provides the high-performance and secure foundation necessary for federal missions.
- Security: AI agents are built with production-ready NVIDIA NIM™ microservices. These NIM are built from a trusted, STIG-ready base layer and support FIPS mode encryption, making them the essential, pre-validated building blocks for achieving a FedRAMP deployment quickly and securely.
DataRobot provides an NVIDIA NIM gallery, which enables rapid consumption of accelerated AI models across multiple modalities and domains, including LLM, VLM, CV, embedding, and more, and direct integration into agentic AI solutions that can be deployed anywhere. - Reasoning: The agent’s core intelligence is powered by NVIDIA Nemotron models. These AI models with open weights, datasets, and recipes, combined with leading efficiency and accuracy, provide the high-level reasoning and planning capabilities for the agent, enabling it to excel at complex reasoning and instruction-following. It goes beyond simple lookups to connect complex data points, delivering true intelligence, not just data retrieval.
- Speech & Translation: NVIDIA Riva Speech and Translation, enables real-time speech recognition, translation, and synthesis directly at the edge. By deploying Riva alongside AIR-T and AirStack Edge, audio content extracted from RF signals can be transcribed and translated on-device with low latency. This capability allows SIGINT agents to turn intercepted voice traffic into actionable, multilingual data streams that seamlessly flow into DataRobot’s agentic AI workflows.
A collaborative approach to mission-critical AI
The combined strengths of DataRobot, NVIDIA, and Deepwave create a comprehensive, secure, production-ready solution:
- DataRobot: End-to-end AI lifecycle orchestration and control.
- NVIDIA: Aaccelerated GPU infrastructure, optimized software frameworks, validated designs, secure and performant foundation models and microservices.
- Deepwave: RF sensors with embedded GPU edge processing, secure datalinks, and streamlined orchestration software.
Together, these capabilities power the Radio Intelligence Agent solution, demonstrating how agentic AI, built on the DataRobot Agent Workforce Platform, can bring real-time intelligence to the edge. The result is a trusted, production-ready path to data sovereignty and autonomous, proactive intelligence for the federal mission.
For more information on using RIA to turn RF data into real time insights, visit deepwave.ai/ria.
To learn more about how we can help advance your agency’s AI ambitions, connect with DataRobot federal experts.
The post Mission-ready AI: Radio intelligence at the edge appeared first on DataRobot.
This week, we are participating in the IEEE/RSJ International Conference on Intelligent Robots and Systems #IROS2025 in Hangzhou #China 
(IRI researchers right-left): @juliaborrassol.bsky.social, David Blanco-Mulero and Anais Garrell 
The “steerable scene generation” system creates digital scenes of things like kitchens, living rooms, and restaurants that engineers can use to simulate lots of real-world robot interactions and scenarios. Image credit: Generative AI image, courtesy of the researchers. See an animated version of the image