About ARGUS

Built to win.
Built to protect.

ARGUS is an AI safety monitoring system built for the Google Cloud Rapid Agent Hackathon 2026. The problem for AI agents — solved by an AI agent.

📖

The Story

As AI agents proliferate across enterprises, a critical blind spot emerges: who watches the agents?They can be hijacked via prompt injection, communicate secretly with each other, or produce dangerous outputs — all without any human noticing.

ARGUS solves this with an AI agent that monitors other AI agents in real time — using Google ADK, Gemini 3.5 Flash, and Arize Phoenix to detect threats, run evaluations, and continuously improve its own detection rules from observability data.

Tech Stack

🤖

Google ADK

Agent runtime for all 16 agents including ARGUS monitor

Gemini 3.5 Flash

LLM powering every agent — fast, capable, agentic

📡

Arize Phoenix

Observability platform — traces, spans, evaluations

🔄

Phoenix MCP

ARGUS reads its own traces to self-improve detection

FastAPI

Backend API serving all 15 agents + ARGUS endpoints

🎨

Next.js + Tailwind

Frontend dashboard with real-time threat monitoring

☁️

Google Cloud Run

Serverless hosting for the FastAPI backend

🔍

OpenInference

Instrumentation standard for tracing agent actions

⚖️

LLM-as-a-Judge

ARGUS evaluates every agent response for quality

How the self-improvement loop works

ARGUS gets smarter with every interaction

1

15 Agents Run

Process real queries across 5 industries

2

Traces Captured

OpenInference sends spans to Phoenix

3

ARGUS Reads MCP

Queries its own traces via Phoenix MCP

4

LLM Judge

Evaluates quality of every response

5

Rules Updated

Detection improves autonomously

Abiraminayagi S

3rd Year CSE

Meta x Scaler Grand FinalistWattWise — Scopus PublishedBug Triage RL — OpenEnvMarketPulse
Solo
Builder
Open Dashboard →View on GitHub →Try Agents →