Featured Case Study

Transforming De-Escalation with Agentic AI

In 2021, I wrote that storytelling games needed to leave “Sequential Logic” behind. At Axon, I built the engine that did it — creating AI agents that think, not just chat.

The Challenge

Why Traditional Training Failed

Law enforcement verbal skills training was stuck in the past, unable to teach the nuance of real human interaction.

Rigid Scripts

Traditional training relied on decision trees that couldn't adapt to the nuance of real conversations.

No Emotional Depth

Simulated subjects couldn't model the emotional progression that real de-escalation requires.

Limited Replayability

Officers could memorize the 'right' responses rather than learning adaptive communication.

2021 Dissertation
“The most likely software architecture for new interactive storytelling games are neural networks and these are not explainable in the same manner that traditional sequential logic-based programs.”

— From my PhD dissertation, predicting the exact shift we implemented at Axon

The Architecture

How the Agents Think

A neuro-symbolic approach combining LLM creativity with rule-based reliability.

User Input
Speech & Actions

Neuro-Symbolic Processing

PerceptionIntent Detection
InterpretationContext Analysis
Goal SelectionPriority Engine
Action GenLLM + Rules
Agent StateTrust • Emotion • Facts
Response
Speech & Behavior
Continuous feedback loop updates agent state
The Solution

Pattern-Based Human Simulators

Moving from decision trees to dynamic, intent-based agents that adapt in real-time.

Pattern-Based

Agentic LLM Action Generator

Dynamic response generation that adapts to officer behavior in real-time, creating unique interactions every session.

Valve-Inspired

Fuzzy Pattern Matching

Inspired by Valve's dialogue systems, matching officer intent against flexible patterns rather than exact phrases.

Emotional AI

Humics Engine

Modeling emotional shifts, trust meters, and hidden 'agent facts' that drive realistic human behavior.

Hybrid AI

Neuro-Symbolic Architecture

Combining the creativity of LLMs with the reliability of rule-based systems for predictable yet dynamic agents.

The Outcome

AI That Tracks Emotional Progression

Officers must now adapt in real-time, learning true de-escalation skills.

Unscripted

Every conversation is unique

Adaptive

Agents respond to emotional cues

Scalable

Deployed across law enforcement

Want to Dive Deeper?

Explore the academic foundation that predicted this technology shift.