Art of Novel Signals: Predicting and Forecasting with High Confidence

This Department of Defense SBIR solicitation, solicitation number DPA26BZ04-DV015, seeks Direct to Phase II proposals to develop a predictive geopolitical forecasting model utilizing temporal knowledge graphs and in-context learning. The objective is to increase event-forecasting timelines from days to weeks while improving precision to at least 90% in data-sparse, denied environments. The project addresses the "data wall" in predictive AI by harvesting untapped, novel signals from multilingual radio, local news, and community audio. System development involves four key areas: establishing a radio data engine, fine-tuning automatic speech recognition (ASR) for predominantly oral languages, implementing a cost-effective synthetic data strategy, and integrating forecasting models. The geographical areas of interest include Central and Southeast Asia, East and Northeast Africa, and South America. Proposers must bypass Phase I by demonstrating a viable theoretical framework and a working, multi-modal in-context learning pipeline that achieves 80% baseline precision. Phase II spans an 18-month base period with sequential milestones to expand language coverage, measure accuracy, improve prediction horizons, and integrate paralinguistic audio features (e.g., pitch and emotional arousal). A $500,000, six-month option period will focus on operational transition and rapid language onboarding. Phase III aims for transition into the Department of Defense, intelligence community, and private sector markets. Dual-use applications include Combatant Command threat warning, humanitarian disaster planning, and political risk monitoring for commercial supply chains.