VoxCroft combines intelligence tradecraft honed through experience with cutting edge AI tools.
VoxCroft’s proprietary intelligence tradecraft, called “population-centric intelligence,” is a methodology that combines hard-to-obtain data sources, state of the art machine translation, AI/ML analytic tools for public sentiment and malign influence detection
Highly differentiated multimedia datasets verified by human crowds for accuracy and community-level context
Industry-leading translation models yield real-time insight and circumvent the need for expensive pools of cleared linguists
Proprietary algorithms discern public opinion, malign influence, and other dangerous societal dynamics portending instability
Highly differentiated multimedia datasets verified by human crowds for accuracy and community-level context
High-quality and relevant data is central to VoxCroft’s media monitoring and data analytics platform solution. VoxCroft’s data collection focuses on media sources and social media platforms carrying content on the topics, issues, and actors most relevant to priorities, not what is easiest to collect. Each region has a unique information environment, where public discourse and information flows across a unique set of traditional media vectors and social media platforms. In turn, we cast a broad net across all media platforms to ensure that we have a complete picture of the information environment, including:
Industry-leading translation models yield real-time insight and circumvent the need for expensive pools of cleared linguists
After collecting millions of data points across platforms, languages, countries, and cultures, Arrow can quickly translate data for analytic exploitation. We use our own proprietary language translation models to process data collected in niche languages, as well as commercially available software if such models perform well. VoxCroft’s translation models are built on high-quality datasets we populate ourselves using our proven track record in gathering, tasking, and managing crowds of language specialists across the world. Indeed, such capabilities are the only commercially available translation models and datasets for national security purposes. We have developed our own industry-leading translation models for low-resource languages around the world to maintain situational awareness in those critical markets.
VoxCroft combines speech-to-text translation and generative AI models to provide English summaries of vernacular language radio broadcast content. Such a capability is critical to gaining the fastest insights into high-priority media reports and events from low-resource languages across any region.
Proprietary algorithms discern public opinion, malign influence, and other dangerous societal dynamics portending instability
Once our data has been collected and translated into a usable format, we run our datasets through a combination of AI models and machine-learning algorithms to tell a compelling story about informational threats and influence in the information environment. By combining the speed and reliability of machine models with high-quality data and translations, we can glean insights that are both targeted and broad, even from massive datasets.
Theme classification models will track and visualize the prominent topics of media discourse and the issues shaping public attitudes toward specific brands and other important actors.
VoxCroft “X-Sentiment” models measure media and public attitudes towards specific actors over time. “Emotions” models further classify sentiment data by common emotions, such as anger and fear.
A mix of AI models discern fractions of discourse driven by bots and other bad faith actors, as well as content containing mis/dis/maliformation to assess the origins and impact of malign influence.
Prominent voices and networks are shown to identify the key networks for targeted strategic communications. AI models also provide geographic and psychographic target audience segmentation.