An AI Curriculum Engine Cuts 9-1-1 Dispatcher Training Reviews to Seconds, Not Minutes

Metro Nashville's emergency communications department has a personnel problem shared by nearly every 911 center in the country: training new call-takers to competence across more than a thousand interdependent skills, each tied to a different incident type or protocol nuance, takes time that training officers rarely have to spare. A new system called PACE, the Personalized Adaptive Curriculum Engine, was built specifically to close that gap, and the research behind it has now been accepted at IJCAI-ECAI 2026, one of the field's most competitive AI conferences.

The paper, first posted to arXiv in March 2026 and revised in July, describes PACE as a system developed in direct partnership with the Metro Nashville Department of Emergency Communications. Rather than following a fixed lesson plan, PACE tracks each trainee as a probabilistic belief over their current skill state, continuously updating estimates of what they know, what they've forgotten, and where they're likely to struggle next. It borrows from cognitive science models of learning and forgetting to decide not just what a trainee should study, but when they need to revisit material they've already covered before it decays.

How the engine decides what to teach next

PACE relies on a structured skill graph that maps how competencies depend on one another, letting it move through diagnostic assessment far faster than a human instructor manually cataloguing a trainee's gaps. On top of that graph, it applies contextual bandit algorithms, a class of methods built for sequential decision-making under uncertainty, to select training scenarios that target skills the trainee is actually ready to absorb rather than presenting problems that are either redundant or premature.

The performance numbers are the paper's headline claim. In studies pairing the AI system against practicing training officers on real-world cases, PACE's pedagogical judgments aligned with expert human judgment 95.45% of the time. On raw efficiency, the system cut the turnaround time for assessing a trainee's progress and recommending next steps from 11.58 minutes down to 34 seconds, a 95.08% reduction that matters enormously in understaffed dispatch centers where training officers are also fielding operational duties.

Against other state-of-the-art adaptive learning frameworks, PACE trainees reached competence 19.50% faster and ended up with 10.95% higher terminal mastery scores, according to the paper's empirical results. Those two figures matter differently: speed addresses the staffing shortages that plague 911 centers nationally, while the mastery gain speaks to whether faster training actually holds up.

The IJCAI-ECAI acceptance places PACE among peer-reviewed work rather than an internal vendor claim, which carries weight in a domain where errors in training translate directly into real-world dispatch mistakes. No conference dates or venue details for the 2026 event have been disclosed alongside the paper, and the research itself stops short of describing any rollout beyond the Metro Nashville pilot. Whether the underlying skill-graph and bandit approach generalizes to other high-stakes, protocol-heavy training domains, such as air traffic control or nursing triage, remains an open question the authors don't directly address.