CONCORDIA by TACITUS

AI-Powered Live Mediation Platform

Created by Giulio Catanzariti — Team TACITUS.me

---

What is CONCORDIA?

CONCORDIA is a real-time AI mediation platform that guides two parties through structured conflict resolution using live audio conversation. An AI mediator listens, speaks, and adapts in real time — extracting psychological indicators, mapping conflict primitives, identifying common ground, and proposing actionable resolution pathways as the parties talk.

The system is built on the TACITUS ontology for conflict representation, combining six structured primitive types (Claims, Interests, Constraints, Leverage, Commitments, Events) with real-time psychological profiling to give mediators and parties a live, evolving picture of the dispute.

How It Works

Live Mediation Session

Two parties sit in front of the app. The AI mediator guides a structured conversation through six phases:

1. Opening — Welcome, ground rules, introductions
2. Discovery — Individual statements from each party (one at a time)
3. Exploration — Cross-referencing perspectives, identifying patterns and triggers
4. Negotiation — Brainstorming options, exploring flexibility
5. Resolution — Narrowing down viable pathways, testing agreements
6. Agreement — Summarizing outcomes, confirming next steps

Real-Time Intelligence

As the conversation progresses, the system provides:

• Psychological Profiles — Emotional state, engagement level, communication style, cooperativeness, defensiveness, key needs, and risk factors for each party
• Case Structure — Claims, Interests, Constraints, Leverage, Commitments, and Events extracted and attributed to each party
• Common Ground & Tension Points — Automatically identified areas of agreement and disagreement
• Resolution Pathways — Concrete proposals with trade-off analysis drawn from established mediation frameworks

Multi-Agent Architecture

Agent	Role
Listener	Real-time voice I/O via Gemini Live Audio API. Maintains conversational flow, calls tools to update the workspace UI.
Profiler	Assesses psychological indicators and emotional state of each party in real time.
Extractor	Parses transcripts into structured JSON primitives (TACITUS ontology).
Advisor	Analyzes the case graph to generate pathways, critical questions, and common ground.

Additional Tools

• Advisor Chat — Ask the Advisor Agent strategic questions about any conflict
• Audio Transcription — Record and transcribe audio using Gemini
• Speech Engine — Generate natural speech from text for advisor responses
• Resolution Library — Reference established mediation frameworks (Principled Negotiation, Transformative, Narrative)

Architecture

┌─────────────────────────────────────┐
│           React Frontend            │
│  (Workspace, Chat, TTS, Transcribe) │
│         ↕ REST + WebSocket          │
├─────────────────────────────────────┤
│         Express.js Backend          │
│   API routes + WebSocket proxy      │
│         ↕ Vertex AI SDK             │
├─────────────────────────────────────┤
│      Google Cloud Vertex AI         │
│  Gemini Live Audio, Flash, TTS      │
└─────────────────────────────────────┘

All AI calls are server-side. The frontend communicates with the backend via REST API (/api/*) and WebSocket (/api/live for live audio sessions). Credentials never reach the browser.

Tech Stack

• Frontend: React 19, TypeScript, Tailwind CSS 4, Framer Motion
• Backend: Express.js with WebSocket support (ws)
• AI: Google Gemini via Vertex AI (Live Audio, Flash, TTS)
• Deployment: Docker, Google Cloud Run
• State: Client-side with localStorage persistence

Run Locally

Prerequisites: Node.js 18+, a Google Cloud service account with Vertex AI API enabled.

npm install

Create a .env.local file:

# Paste your full service account JSON (single line)
GOOGLE_SERVICE_ACCOUNT_JSON='{"type":"service_account","project_id":"your-project",...}'
GOOGLE_CLOUD_PROJECT=your-project-id
GOOGLE_CLOUD_LOCATION=us-central1

Start the backend server (required for API calls):

npx tsx server.ts

In a separate terminal, start the frontend dev server:

npm run dev

Open http://localhost:3000. The Vite dev server proxies /api/* requests to the backend on port 8080.

Deploy with Docker

docker build -t concordia .

docker run -p 8080:8080 \
  -e GOOGLE_SERVICE_ACCOUNT_JSON='{"type":"service_account",...}' \
  -e GOOGLE_CLOUD_PROJECT=your-project-id \
  -e GOOGLE_CLOUD_LOCATION=us-central1 \
  concordia

Deploy to Google Cloud Run

# Build and push
gcloud builds submit --tag gcr.io/YOUR_PROJECT/concordia

# Deploy (pass credentials as env var)
gcloud run deploy concordia \
  --image gcr.io/YOUR_PROJECT/concordia \
  --platform managed \
  --region us-central1 \
  --allow-unauthenticated \
  --set-env-vars "GOOGLE_CLOUD_PROJECT=YOUR_PROJECT,GOOGLE_CLOUD_LOCATION=us-central1" \
  --set-secrets "GOOGLE_SERVICE_ACCOUNT_JSON=concordia-sa-key:latest"

Environment Variables

Variable	Required	Description
GOOGLE_SERVICE_ACCOUNT_JSON	Yes*	Full JSON content of your GCP service account key
GOOGLE_CLOUD_PROJECT	Yes*	Google Cloud project ID
GOOGLE_CLOUD_LOCATION	No	Vertex AI region (default: us-central1)
USE_VERTEX_AI	No	Set to false to use a Gemini API key instead
GEMINI_API_KEY	No	Required only if USE_VERTEX_AI=false
MODEL_LIVE	No	Live audio model (default: gemini-2.0-flash-live-001)
MODEL_TEXT	No	Text/chat model (default: gemini-2.0-flash)
MODEL_TTS	No	TTS model (default: gemini-2.5-flash-preview-tts)
MODEL_TRANSCRIBE	No	Transcription model (default: gemini-2.0-flash)

* Not required if USE_VERTEX_AI=false and GEMINI_API_KEY is provided.

License

Apache-2.0