Skip to content

sambhandavale/arbiter

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

14 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Frame 35241

ARBITER: Multi-Agent Supply Chain Contradiction Resolver

"Don't just predict the future. Cross-examine it."

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.


🛑 1. The Problem

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.

💡 2. The Unique Solution: Adversarial Consensus

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.

🛠️ 3. Innovative Use of Google Technology

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.

🏗️ 4. System Architecture

ARBITER utilizes a Sequential Multi-Agent Pipeline. Every query is pressure-tested through five layers of specialized intelligence.

The Agent "Round Table"

  1. 🕵️ Extraction Specialist: Standardizes the user's claim into structured logistics keys (Ports, Dates, Suppliers).
  2. 📜 Historian Agent: Queries local JSON performance databases to find the supplier’s true historical reliability.
  3. ⛈️ Weather Agent: Samples waypoints along the route to check real-time telemetry via Open-Meteo.
  4. 🚢 Logistics Agent: Scours the web via SerpApi for current port congestion, vessel waiting times, and terminal disruptions.
  5. ⚖️ Confidence Agent: The final Arbiter. It synthesizes the debate, identifies core contradictions, and issues a Probabilistic Confidence Score.

Screenshot 2026-01-25 081914

💻 5. Tech Stack

  • 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).

🚀 6. Project Setup & Run Steps

Prerequisites

  • Python 3.11+ and Node.js 18+
  • Google AI Studio API Key
  • SerpApi Key

Step 1: Backend Setup

cd backend
uv venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
uv pip install -r requirements.txt

Step 2: Environment Configuration

Create a .env file in the backend/ directory:

GOOGLE_API_KEY=your_gemini_key
SERPAPI_API_KEY=your_serpapi_key

Step 3: Run the API Server

uv run adk api_server --port 8000

Step 4: Frontend Setup

cd frontend
npm install
npm run dev

🔮 7. Future Scope & Scalability

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.

About

ARBITER is an adversarial multi-agent platform powered by Google ADK and Gemini that cross-examines supply chain claims against real-time telemetry. The system provides fact-grounded confidence scores and transparent transcripts to help B2B enterprises proactively mitigate high-stakes risks in logistics, finance, and healthcare.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Contributors