ARBITER is an adversarial AI decision-support system that resolves contradictions between supplier commitments and real-world logistics volatility. Built on the principle of Adversarial Consensus, it utilizes a "Round Table" of autonomous agents to challenge optimistic claims, surface hidden risks, and provide a probabilistic confidence score for every shipment.
Global supply chains are increasingly fragile, yet procurement decisions still rely on static data. Industry reports show:
- Optimism Bias: Supplier-provided delivery guarantees often deviate from reality by 20–40%.
- Invisible Bottlenecks: Port congestion and infrastructure failures can increase lead times by 3x with zero advance notice.
- Signal Collapse: Traditional AI "averages" conflicting data, often suppressing critical minority risks like a localized storm surge or a brewing labor strike.
ARBITER introduces a paradigm shift: Structured Disagreement. Unlike traditional systems that force a single output, ARBITER encourages its agents to disagree.
By preserving these contradictions rather than averaging them away, the system reflects the true volatility of global trade. Resilience emerges from critical challenge, ensuring that high-risk signals are never "smoothed out" of the final verdict.
ARBITER leverages the full power of the Google AI Ecosystem to enable real-time adversarial reasoning:
- Gemini 2.0 Flash: Acts as the high-speed reasoning core for every agent, enabling complex tool use and nuanced debate.
- Google ADK (Agent Development Kit): Orchestrates the sequential "Round Table" logic, managing shared session state across five specialized agents.
- Google Search Grounding: Powers the Logistics Agent to pull real-time maritime intelligence and port congestion metrics directly from the live web.
ARBITER utilizes a Sequential Multi-Agent Pipeline. Every query is pressure-tested through five layers of specialized intelligence.
- 🕵️ Extraction Specialist: Standardizes the user's claim into structured logistics keys (Ports, Dates, Suppliers).
- 📜 Historian Agent: Queries local JSON performance databases to find the supplier’s true historical reliability.
- ⛈️ Weather Agent: Samples waypoints along the route to check real-time telemetry via Open-Meteo.
- 🚢 Logistics Agent: Scours the web via SerpApi for current port congestion, vessel waiting times, and terminal disruptions.
- ⚖️ Confidence Agent: The final Arbiter. It synthesizes the debate, identifies core contradictions, and issues a Probabilistic Confidence Score.
- Backend: Python 3.11+, FastAPI, Google ADK.
- AI Engine: Gemini 2.0 Flash.
- Frontend: Next.js 14, Tailwind CSS, Framer Motion (for real-time debate visualization).
- APIs: SerpApi (Google Search), Open-Meteo (Weather Telemetry).
- Database: Local JSON (Supplier Performance) + SQLite (Session History).
- Python 3.11+ and Node.js 18+
- Google AI Studio API Key
- SerpApi Key
cd backend
uv venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
uv pip install -r requirements.txt
Create a .env file in the backend/ directory:
GOOGLE_API_KEY=your_gemini_key
SERPAPI_API_KEY=your_serpapi_key
uv run adk api_server --port 8000
cd frontend
npm install
npm run dev
The ARBITER architecture is designed for massive horizontal scaling:
- Maritime 2.0: Integration with live AIS (Automatic Identification System) for real-time vessel tracking.
- AgriTech Expansion: Swapping logistics agents for "Crop Yield" and "Soil Sensor" agents to evaluate food supply stability.
- HealthTech Risk: Using adversarial logic to cross-examine diagnostic data vs. pharmaceutical supply availability.
- Operational Stress Profiling: Implementing a feedback loop that "rewards" agents whose skepticism most accurately predicted a real-world delay.