paX | Devpost

Inspiration

Our project, paX, grew out of the Peacegaming framework pioneered at the University of Cambridge, which uses gaming and simulation to model conflict and peacebuilding. It aims to foster empathy and understanding in conflict situations. Peacegaming emphasizes that true security is not only about deterrence but also about dialogue, perspective-taking, and developing trust between diverse actors. That approach showed us how technology can also be used for building empathy skills that are essential in today’s complex operating environments.

For us, this project is rooted in lived experience. One teammate worked with the U.S. Air Force this summer and saw firsthand how empathy, trust, and local relationships can determine the success or failure of initiatives. All of us also have been involved in ongoing conversations around AI bias, cultural sensitivity, and fairness in AI systems. We share a desire to create technology that reflects human complexity, and we strongly believe that empathy is not a “soft skill,” it is mission-critical.

paX became our outlet to bring these experiences together and to create a simulation where AI archetypes allow active duty officers to practice navigating the human dimensions of conflict.

What it does

paX is a peace gaming simulation with AI archetypes that train empathy in U.S. military officers by modeling civil-military interactions. Players engage with archetypes representing local civilians, leaders, NGOs, and journalists in contested regions, experiencing how their choices build or erode trust.

Key features:

AI Archetypes: NPCs such as tribal elders, mayors, NGO workers, women, youth, and journalists who react dynamically to player decisions.
Regional Scenarios: Choose a geopolitical hotspot (e.g., Cambodia, Ukraine, Indo-Pacific) to simulate U.S. soft power competition against adversaries.
Dialogue Simulation: Branching conversations where players interact with regional AI archetypes and receive context-aware NPC reactions.
Empathy & Trust Meters: Real-time and end-of-session feedback on the effects of player actions.
Multilingual Support: Archetypes simulate authentic conversations with civilians across cultural and linguistic contexts.
Adaptive Reasoning: Archetype responses reflect political dynamics, historical mistrust, and adversary influence operations.

How we built it

Building paX meant stitching together very different technologies into one continuous pipeline. Our solution runs on the following framework:

Dialogue Generation: Cohere

Generates adaptive dialogue lines for each archetype. - Grounded in scenario context (e.g., tariffs, Chinese infrastructure, community concerns).
Adjusts NPC tone based on user choices (empathetic, skeptical, or collaborative).

Voice Layer: ElevenLabs

Converts Cohere’s text into natural, Afrikaans-accented English voices.
Tuned for pacing, intonation, and slight structural differences in sentence rhythm.
Exported as MP3/WAV for smooth syncing.

Visual Archetypes: HeyGen

Animates avatars with realistic facial and lip-syncing tied to ElevenLabs audio.
Archetypes designed to reflect local community leaders, officials, and activists.
Provides players with a face-to-face experience instead of plain text.

User Interface: React + Framer

Interactive PeaceGaming interface with clickable dialogue options.
Side-by-side layout: player choices on the left, avatar video on the right, and empathy/trust meters along the bottom.
Empathy meter updates dynamically with each decision.

Metrics & Feedback

Prototype tracks decision impact through trust/empathy shifts.
Future integration planned for more structured metrics (ethical agility, adaptability, decision rationale).

Challenges we ran into

One of the hardest parts was cultural sensitivity. We didn’t want our archetypes to feel like clichés. They needed to be authentic, but also safe and respectful. That balance was tough.

We also had to pull together a lot of sponsor technologies in a very short time. APIs don’t always play nicely, and we spent late nights making them talk to each other.

The design of the interface was another challenge. We knew we couldn’t build out every feature in a weekend, so we made hard choices. Keep it simple. Focus on what tells the story.

Accomplishments that we're proud of

Over the past weekend, we built something that could go far beyond a simple demo: a simulation that captures the tension, nuance, and human dimension of global security challenges. By grounding our archetypes in real-world issues...the South African trade disputes, Chinese soft power, and civilian–military interactions... we learned how AI can make training more realistic and more empathetic.

What we’re most proud of, though, is how deliberate we were in the ideation process and building a plan for a working prototype. We took time at the start to work though ideas, our motivations as a team, and what inspires us. That led each of us to paX, and we couldn't be more grateful. In less than two days, we've built a system where users can sit across from an AI community leader, be challenged, and practice responding with empathy. For us, that combination of urgency, creativity, and collaboration is the real win.

What we learned

This project pushed us to think about AI in ways that go beyond raw capability. We learned how important it is to design systems that don’t just generate content, but capture tone, empathy, and cultural nuance. Small differences in phrasing or cadence can completely change whether a conversation builds trust or creates tension.

Technically, we stretched ourselves by integrating very different tools into one experience. Cohere forced us to structure prompts that could handle context like tariffs and foreign influence without falling back on generic answers. ElevenLabs taught us how to tweak outputs for accent and pacing to sound authentic, and MetaHuman/HeyGen showed us how to turn static text into a human presence that people can actually respond to.

Most of all, we learned that realism requires iteration. We spent time revising dialogue, testing voices, and adjusting our UI to make sure the interaction felt grounded instead of gimmicky. That process gave us a deeper appreciation for how empathy and technical design go hand in hand in creating something believable.

What's next for paX

Our next steps are all about making paX more immersive and technically robust. On the feature side, we want to:

Expand scenarios across regions like Africa, the Indo-Pacific, and Eastern Europe, each with archetypes tied to current affairs.
Auto-populate archetype knowledge bases with live news so dialogues evolve as world events change.
Add multilingual, voice-driven dialogue with speech-to-text and text-to-speech, so players can interact naturally.
Introduce more nuanced trust and empathy meters that shift dynamically with player choices.
Build an after-action insights dashboard to help users reflect on their biases, blind spots, and decision patterns.
Develop a scoring system that adapts to tone and outcomes, so players can get immediate feedback on how their interactions affect trust.

Once these features are in place, we would love to see paX piloted to test usability and refine training value. Over the long term, we see paX reshaping the United States's soft power influence by becoming a scalable interface for Professional Military Education to use and help prepare leaders to navigate the human side of conflict.

We see this prototype as the start of a much larger platform. Our vision for paX is a training tool that blends policy relevance, cultural realism, and AI technology into one experience...a platform that could help military and civilian leaders practice the hardest part of strategy: understanding people.