Jared Alexander Dumale

Early-warning work in intelligence shops slows down at one point: when an analyst has to read raw text, pull out the key details, and place them on a map. I wanted to know whether a small pipeline built on a large language model could take on that work and hand back something an analyst could actually use.

I built a proof of concept that took in open-source reporting, used a language model to pull out names, organizations, and locations, wrote structured event summaries, and sent the geocoded results to a GIS dashboard that showed trends. The pipeline ran start to finish with no manual cleanup.

Twelve demo sessions with analysts gave concrete feedback on accuracy, presentation, and how much to trust the output. That is the kind of input that turns a research prototype into a tool people will use. The work shows how applied GeoAI can shorten the time between a signal and a decision.

USAFA cadets need to know which countries fall under which Combatant Command. Flashcards and slides do not make it stick, and the test format rewards short-term memorization over real geographic intuition.

Geo412 turns the problem into a game. It is a browser-based platform with several trivia modes built around the world map, made for self-paced study that rewards quick recognition and pattern-spotting. Cadets play in short sessions, and the geography of the answers reinforces itself.

The project shows that a small, focused web tool can beat official training material when it is designed around how people actually learn. It is live and in use, and it shows how curriculum design and lightweight development strengthen each other. The platform also supports my research into how AI can help design learning aids and effective teaching tools.

COVEY 535 was a USAF aircraft lost over Laos. Decades later, recovery teams still need a defensible way to decide where to search first, and the data they need is scattered across historical imagery, terrain models, and archived mission records.

I led an undergraduate research team that brought 22 historical imagery scenes and terrain analyses from digital elevation models into one geospatial workflow. The result was four map products that narrowed the priority search zones, built from three different data types and developed with three partner institutions.

The project shows how geospatial analysis can serve a humanitarian mission and build student skills at the same time. It is a case study in combining mentorship, multi-source data fusion, and decision-support mapping on a problem with a real outcome.

Career planning at USAFA depends on understanding the training pipeline for each Air Force Specialty Code (AFSC): where you go, when, and for how long. That information is scattered across PDFs, briefings, and what individual advisors happen to know.

The AFSC Atlas, a capstone I supervised, pulls that data into a single ArcGIS Dashboard. A cadet enters their AFSC and immediately sees training bases on a map, milestones in order on a timeline, and key durations at a glance. The dashboard was built with Arcade, HTML, CSS, and JavaScript, with the code managed in GitHub. The underlying dataset was cleaned and structured so it can be reused beyond this proof of concept.

Beyond the product itself, the project shows how a small geospatial dashboard can move an advisor and cadet conversation from guesswork to specifics. It also shows that supervising student work can produce tools the academy can actually use.

GEOINT products are often read under time pressure, and how a map or image is styled shapes how quickly and how accurately an analyst can read it. This cadet-led research project tests how changes to visual hierarchy affect the speed and accuracy of GEOINT interpretation.

I serve as associate investigator, guiding the cadets on research design and analysis. The goal is to identify which GEOINT visualizations are best suited for rapid interpretation, so that product design choices rest on evidence rather than habit.