Scaling High-Stakes Conversation Practice with AI
To make difficult conversations trainable through simulation, feedback, and real-time coaching.
Timeline
Sept 2022 to Jan 2024
Role
Learning Experience Designer & Product Owner — Strategy, Research, AI Integration & Delivery
Business Goal
Most organizations have frameworks and resources for navigating difficult conversations. What they lack is a way to practice them at scale. High-stakes dialogue skills (the kind required to bridge differences in identity, perspective, and lived experience) are built through repetition, feedback, and exposure to increasing levels of challenge. Without a scalable way to create those conditions, organizations are left hoping that exposure to content translates into behavior change (spoiler alert: it rarely does). CivicXChange was built to close that gap: an AI-powered platform that shifts learning from content delivery to deliberate skill practice, scaling access to high-quality dialogue training without relying on facilitators.
Problem Statement
Facilitated dialogue programs work, but they don't scale. High-quality facilitation is expensive, logistically complex, and dependent on the skill of the facilitator. Access is uneven and repetition is rare. Participation is typically voluntary, which means the people who most need to build these skills are least likely to show up.
Digital alternatives don't replicate the conditions required for skill development. Static content and pre-scripted scenario pathways cannot replicate the emotional intensity, social dynamics, or power imbalances of real conversations across difference. VR requires significant investment to build, breaks down quickly as discourse evolves, and shows limited evidence of sustained behavior change. No existing solution gives organizations a scalable, repeatable way to build this capability or measure whether it's working.
The 60+ page study documenting platform design, learner data, and evidence of efficacy across 22 participants at six universities.
The instructional storyboard from an early branching scenario prototype: the frame-by-frame design logic, learner states, and learning science rationale that informed the final AI-mediated experience.
Solution
I designed and built CivicXChange as an AI-enabled deliberate practice platform for high-stakes communication, giving learners a repeatable, low-risk environment to build both skill and endurance for navigating conversations across real differences in identity, perspective, and lived experience.
The core architecture is a dual-agent system: each learner engages simultaneously with an AI conversation partner representing an opposing perspective and an AI dialogue coach providing real-time facilitation and feedback. This structure replicates both the tension of genuine disagreement and the guidance of a skilled facilitator, at any scale, without human facilitation overhead.
I developed and iterated across three agile sprints. The first version used two fixed diplomatic personas, which failed to create sufficient challenge. The second expanded to three richly developed personas, improving engagement but remaining limited by predictability. The final iteration eliminated fixed personas entirely, implementing a dynamic generation system that creates a unique opposing character in real time based on the learner's inputs, ensuring every interaction is context-specific, unpredictable, and appropriately challenging.
The experience is structured as a repeatable learning loop: live simulation, guided reflection, and immediate feedback. The dialogue coach continuously adapts to the learner's tone, reasoning, and communication patterns, reinforcing productive behaviors and intervening when conversations escalate or break down. All interactions are grounded in high-stakes topics including race, class, and immigration to ensure the emotional and social realism required for meaningful skill transfer.