2) Co-designed

Core co-design components

• Language packs: modular dictionaries per locale (EN-UK veterans, UA frontline).
• Lexicon registers: grouped phrases by stance (contain, witness, redirect).
• Cultural validation: review by bilingual clinicians and veteran focus groups.
• Iterative loops: anonymised transcripts feed advisory review and refinement.
• Ethical guardrails: supervision for non-pathologising practice and safe escalation.

Co-design loop (diagram)

Veteran & Clinician Input
        |
Lexicon Register <-> Language Pack (EN / UA)
        |
Cultural Vetting <-> UK Research / Ukraine Partners
        |
Prompt Library -> Live Deployment -> Feedback Logs
        ^                                   |
        |-----------------------------------|
               Co-Design Review Loop

Runtime register selection (pseudo)

const register = user.language === "uk" ? "ua_frontline" : "en_veteran";
const phrase = lexicon[register].grounding[0];
assistant.reply(phrase);

Voices and prosody: humanising speech output

Personas receive distinct voices and tuned prosody. Slow ramp-in, micro-pauses, and softer endings reduce cognitive load. Safety language uses calm cadence; values prompts sound warmer.

const voiceMap = {
  charlie: { en: "en-GB-RyanNeural", uk: "uk-UA-OstapNeural" },
  mishka:  { en: "en-GB-SoniaNeural", uk: "uk-UA-PolinaNeural" }
};
const voiceName = voiceMap[persona][user.language];

<prosody rate="-10%" pitch="-1st"> calmer safety lines
<break time="400ms"/> gentle pause
<emphasis level="reduced"> soften corrective statements
<prosody rate="+5%"> action planning (not rushed)

const region = process.env.AZURE_TTS_REGION;
const key = process.env.AZURE_TTS_KEY;
const voice = "en-GB-SoniaNeural";

const ssml = `
<speak version="1.0" xml:lang="en-GB">
  <voice name="${voice}">
    <prosody rate="-5%" pitch="-1st">
      I am here with you. We will take one small step.
    </prosody>
    <break time="300ms"/>
    <prosody rate="-5%" pitch="-1st">
      If things feel intense, we can pause.
    </prosody>
  </voice>
</speak>`;

const res = await fetch(`https://${region}.tts.speech.microsoft.com/cognitiveservices/v1`, {
  method: "POST",
  headers: {
    "Ocp-Apim-Subscription-Key": key,
    "Content-Type": "application/ssml+xml",
    "X-Microsoft-OutputFormat": "audio-16khz-32kbitrate-mono-mp3",
  },
  body: ssml,
});
const audioBuffer = await res.arrayBuffer();