From words to worlds - How AI creates operational scenarios for the Police
How do you prepare for an event that hasn’t happened yet?
Whether it’s a large demonstration, a major public event, or an unexpected security incident, operational teams need to make critical decisions under uncertainty. They need to understand the environment, anticipate how people will move, evaluate potential risks, and assess the impact of different interventions; often within a matter of hours.
Traditionally, this preparation relies on maps, expert knowledge, and lengthy planning sessions. Building realistic simulation scenarios typically requires specialist software and considerable technical expertise.
But what if you could simply describe a situation in natural language, and let AI do the rest?
That question became the foundation of PREP (Police Rapid Environment and Preparation), an innovation project developed by uCrowds during Phase 1 of the Innovation Impact Challenge for the Dutch Police.
Turning natural language into operational scenarios
Imagine typing or speaking a simple instruction:
“Simulate a demonstration at the city square where part of the crowd moves toward the train station while one of the access roads is closed.”
Within minutes, PREP transforms this description into a fully interactive operational scenario.
Behind the scenes, a Large Language Model interprets the request, fills in missing information through a short dialogue if necessary, automatically reconstructs a digital twin of the environment, and connects the scenario to a real-time crowd simulation. Operational teams can immediately explore multiple “what-if” scenarios and evaluate the effects of different interventions.
PREP is an integrated workflow that combines Generative AI, geospatial data, digital twins, and high-performance simulation into a single operational decision-support platform.
Building on proven technology, focusing on accessibility
PREP builds upon SimCrowds, uCrowds’ market-proven crowd simulation platform.
For years, SimCrowds has supported the planning and safety analysis of large public events, enabling organizations to simulate up to 200,000 pedestrians in real time, analyze crowd flows, identify bottlenecks, and evaluate evacuation strategies. It has been used for events such as SAIL Amsterdam, National Remembrance Day, King’s Day, and many more.
Creating detailed operational scenarios still required simulation expertise and significant preparation time. PREP removes that barrier by introducing an AI layer that allows operational personnel to create sophisticated scenarios using natural language instead of specialist simulation tools.
A five-step AI pipeline
The PREP workflow consists of five connected stages:
- Describe the situation. The user explains the scenario using natural language.
- AI understands the request. A locally running Large Language Model interprets the description and asks follow-up questions when necessary.
- Generate the environment. Public geospatial datasets automatically create a semantic 3D digital twin of the location.
- Simulate human behaviour. SimCrowds generates realistic pedestrian movement and crowd dynamics in real time.
- Evaluate and compare scenarios. Users test interventions, compare alternative strategies, and immediately visualize the operational consequences.
Instead of spending hours preparing a simulation, operational personnel can focus on exploring options and making informed decisions.
What we have learned
One of the most valuable aspects of the project was the close collaboration with stakeholders from the Dutch Police.
Initially, PREP was envisioned primarily as a tool for immersive training.
However, interviews revealed an even stronger opportunity.
Operational teams consistently emphasized the need for rapid operational preparation and fast scenario comparison. Rather than replacing existing training programs, PREP can help teams quickly build situational awareness before an operation starts.
Another key insight was the importance of comparing multiple “what-if” scenarios. Decision makers want to understand not only the current situation, but also how different interventions (closing an entrance, redirecting visitors, or deploying additional personnel) change the outcome.
Beyond policing
Although PREP was designed to support the Dutch Police, the underlying technology has an even broader impact. Any organization responsible for managing large numbers of people can benefit from AI-powered operational planning.
Examples include:
- Public safety agencies
- Emergency services
- Municipalities
- Event organizers
- Critical infrastructure operators
- Transportation authorities
The technology also has strong dual-use potential. Defense organizations face similar challenges: rapidly understanding unfamiliar environments, planning operations, preparing personnel, and evaluating complex scenarios. AI-generated digital twins combined with large-scale crowd simulation can significantly reduce preparation time while improving operational insight.
Looking ahead
Phase 1 demonstrated that the concept is technically feasible.
A first proof of concept showed that a digital environment and a simple AI-generated crowd simulation can be created within approximately 10 minutes.
This is only the beginning.
As generative AI continues to evolve, we believe the future of operational planning will become increasingly conversational. Instead of manually constructing complex simulations, professionals will simply describe a situation, ask questions, compare alternatives, and receive immediate insights.
Privacy, transparency, and reliability are core requirements for uCrowds. The AI model runs locally on the user’s own equipment, so no operational data has to leave the organization. Users decide which model to use and can inspect how it reached a given output. And the domain expert remains in charge at every step: the AI’s role is to support the decision, not to make it.
At uCrowds, we are excited to contribute to that future by combining Artificial Intelligence, Digital Twins, and Predictive Human Movement Analytics into practical solutions that support better decision-making.
We would like to sincerely thank the Netherlands Enterprise Agency (RVO) for organizing the Innovation Impact Challenge and supporting innovation within the Dutch public sector. In particular, we thank Nicoline Breed and Robert van Haaften for their guidance and enthusiasm throughout the project, as well as Dominique van Ratingen from CIIIC. We also extend our gratitude to all stakeholders and operational experts from the Politie Nederland who generously shared their experience, feedback, and operational knowledge, including Jelte Altena, Corine Laman, Nanco Oudejans, Gerard ten Buuren, Joeri Sterringa and others.
If you’d like to learn more about PREP, AI-powered operational planning, or the possibilities of large-scale crowd simulation, we’d be delighted to connect.
Let’s shape the future of operational decision support – together.
#crowd #simulation #pedestrianDynamics #llm #publicSafety