Spaces:

Benny-Tang
/

aegis

Configuration error

App Files Files Community

aegis / README.md

Benny-Tang

Upload README.md

f0bfaf0 verified 12 days ago

preview code

raw

history blame contribute delete

11.6 kB

🛡️ Aegis — Autonomous Enterprise Crisis Management

AMD Developer Hackathon 2026 | Track 1: AI Agents & Agentic Workflows

Aegis is a fully autonomous 7-agent AI system that monitors global supply chain risks in real time, scrapes live maritime shipping intelligence, predicts disruptions using a hybrid ML forecasting model, and autonomously executes crisis response decisions — before humans can react.

"Aegis protects enterprises from global disruptions by turning real-time chaos into autonomous decisions."

🎬 Live Demo


Live App	http://YOUR_AMD_INSTANCE_IP:8000
Demo Video	https://youtu.be/w_jquBRuLkA
API Docs	http://YOUR_AMD_INSTANCE_IP:8000/docs
GitHub	https://github.com/Benny-Tang/aegis

🌍 Real-World Relevance

Aegis was built as the Strait of Hormuz crisis unfolded in real time (2026). The exact scenario we demo — oil price spike, shipping disruption, geopolitical escalation — is actively happening right now. Our live MarineTraffic scraper pulls real headlines like:

"Trump's Hormuz Blockade Has Deepened A Historic Shipping Crisis"
"Strait of Hormuz Remains Near-Empty With Just A Few Iran Ships Moving"
"Iran War Leaves Seafarers Stranded In The Gulf"
"US Says Navy Intercepted Iran-Linked Vessel in Arabian Sea"

Aegis responds to this real crisis autonomously in 2 seconds.

🧠 How It Works

When a crisis event is detected, Aegis activates a chain of 7 specialized AI agents that observe, think, predict, simulate, decide, alert and act — autonomously.

Crisis Event Detected
        │
        ▼
┌─────────────────────────────────────────────────────┐
│                AEGIS 7-AGENT PIPELINE               │
│                                                     │
│  📡 Signal Agent      → Detects anomalies +         │
│                          scrapes live MarineTraffic  │
│  🧠 Intelligence Agent → Interprets risk context    │
│  📈 Forecast Agent    → ARIMA + XGBoost prediction  │
│  🧩 Simulation Agent  → Runs what-if scenarios      │
│  ⚡ Decision Agent    → Ranks action plan           │
│  🚨 Alert Agent       → Dispatches notifications    │
│  🔄 Execution Agent   → Triggers workflows          │
└─────────────────────────────────────────────────────┘
        │
        ▼
  Autonomous Response in ~2 seconds
  ($1.7M savings identified per crisis event)

🚢 Live MarineTraffic Intelligence

Aegis scrapes real-time shipping intelligence from multiple maritime sources on every crisis trigger:

Source	Type	Status
gCaptain	Maritime news	✅ Live
TradeWinds	Shipping intelligence	✅ Live
MarineTraffic Blog	Vessel tracking news	⚠️ Partial
Reuters	Geopolitical news	⚠️ Auth required

The Signal Agent automatically enriches every crisis analysis with live shipping alerts — tanker reroutes, port closures, war risk insurance changes, navy intercepts.

Tested live — 8 real shipping intelligence items scraped in a single run with 95% CRITICAL confidence classification by Groq LLaMA 3.3 70B.

⚙️ Tech Stack

Layer	Technology
LLM	Groq — LLaMA 3.3 70B Versatile
ML Forecast	XGBoost 2.0 + ARIMA(2,1,2) hybrid ensemble
Marine Intelligence	MarineTraffic + gCaptain + TradeWinds scraper
GPU Acceleration	AMD Instinct MI300X via ROCm 6.2
API	FastAPI + Uvicorn + Gunicorn (4 workers)
Streaming	Server-Sent Events (SSE) — live agent updates
Deployment	AMD Developer Cloud (amd.digitalocean.com)
Frontend	Vanilla JS + CSS — dark terminal aesthetic
Auto-Monitor	Autonomous scan every 5 minutes 24/7

🚀 Quick Start

Option A — Google Colab (easiest)

Open Aegis_Hackathon.ipynb in Google Colab
Fill in your GROQ_API_KEY and NGROK_TOKEN in Cell 1
Click Runtime → Run All
Copy the ngrok URL printed at the bottom

Option B — AMD Developer Cloud (production)

# 1. SSH into your AMD MI300X instance
ssh -i your-key.ppk root@YOUR_AMD_IP

# 2. Set your Groq API key
export GROQ_API_KEY=gsk_YOUR_KEY_HERE

# 3. Install packages
pip3 install fastapi "uvicorn[standard]" groq httpx pydantic \
    numpy pandas scikit-learn xgboost statsmodels \
    python-multipart aiofiles gunicorn \
    requests beautifulsoup4 lxml

# 4. Create project folders
mkdir -p /opt/aegis/agents /opt/aegis/models /opt/aegis/api /opt/aegis/logs

# 5. Upload project files and start
cd /opt/aegis && gunicorn api.server:app \
    --workers 4 \
    --worker-class uvicorn.workers.UvicornWorker \
    --bind 0.0.0.0:8000 \
    --timeout 120 \
    --daemon

# 6. Open in browser
http://YOUR_AMD_IP:8000

📁 Project Structure

aegis/
├── agents/
│   └── swarm.py              # 7 Groq-powered autonomous agents
│                             # + MarineTraffic live scraper
├── models/
│   └── forecaster.py         # ARIMA + XGBoost hybrid ML model
├── api/
│   └── server.py             # FastAPI + SSE streaming server
├── frontend.html             # Live dashboard — dark terminal UI
├── Aegis_Hackathon.ipynb     # Google Colab notebook
├── requirements.txt          # Python dependencies
└── LICENSE                   # MIT License

🤖 The 7 Agents

1. 📡 Signal Agent (Watcher)

Monitors real-time feeds — oil prices, shipping routes, geopolitical news. Automatically scrapes live MarineTraffic, gCaptain and TradeWinds shipping intelligence on every trigger. Classifies anomalies by severity (LOW → CRITICAL) with confidence score.

2. 🧠 Intelligence Agent (Interpreter)

Uses LLaMA 3.3 70B to convert raw signals and marine data into structured risk context. Identifies root causes, affected regions, and escalation probability.

3. 📈 Forecast Agent (Predictor)

Runs a hybrid ARIMA(2,1,2) + XGBoost ensemble model to predict oil prices, logistics delays, and cost impact over a 14-day horizon. XGBoost runs on AMD MI300X GPU via ROCm for accelerated inference.

4. 🧩 Simulation Agent (Strategist)

Generates 3 weighted what-if scenarios (best case, base case, worst case) with probability weights and cost impact estimates for each.

5. ⚡ Decision Agent (Brain)

Synthesizes all upstream intelligence into a ranked action plan with dollar savings estimates, implementation timelines, and risk ratings.

6. 🚨 Alert Agent (Communicator)

Formats and dispatches real-time notifications to Slack, email, and live dashboard. Generates executive summaries for C-suite stakeholders.

7. 🔄 Execution Agent (Operator)

Triggers downstream enterprise workflows autonomously — ERP supplier switches, procurement API calls, logistics rerouting.

📊 ML Forecasting Model

Why ARIMA + XGBoost hybrid?

Component	Role	Weight (crisis)
ARIMA(2,1,2)	Linear trend + autocorrelation	30%
XGBoost (ROCm GPU)	Non-linear geopolitical signals	70%

Features engineered: lag-1, lag-3, lag-7, lag-14, rolling mean/std (7 & 14 day), 3-day % change, day-of-week

AMD MI300X advantage: XGBoost tree building runs 8–15x faster on MI300X vs CPU via ROCm CUDA compatibility layer.

🔌 API Reference

Method	Endpoint	Description
`GET`	`/`	Live dashboard
`GET`	`/health`	System health check
`GET`	`/docs`	Swagger UI
`POST`	`/api/crisis`	Full 7-agent pipeline (JSON)
`GET`	`/api/stream`	SSE live agent stream
`POST`	`/api/forecast`	ML forecast only
`GET`	`/api/marine`	Live MarineTraffic shipping feed
`GET`	`/api/status`	Live system status

Example — trigger crisis

curl -X POST http://YOUR_IP:8000/api/crisis \
  -H "Content-Type: application/json" \
  -d '{
    "oil_price_change_pct": 18.0,
    "shipping_disruption": "Strait of Hormuz — 3 tankers rerouted",
    "news_headline": "Regional conflict escalation near Persian Gulf",
    "severity": "HIGH",
    "disruption_factor": 0.7,
    "horizon_days": 14
  }'

Example — get live marine intelligence

curl http://YOUR_IP:8000/api/marine

Example — SSE stream

const es = new EventSource('http://YOUR_IP:8000/api/stream?oil_change=18&disruption=0.7');
es.onmessage = (e) => {
  const msg = JSON.parse(e.data);
  console.log(msg.type, msg.agent, msg.data);
};

💡 Business Value

Aegis addresses a $1.5 trillion annual problem — global supply chain disruptions caused by geopolitical events, commodity shocks, and logistics failures.

Key metrics from live demo (Strait of Hormuz crisis):

Metric	Value
Pipeline response time	2 seconds
Agents active simultaneously	7 / 7
Savings identified	$1.7M per crisis
Logistics delay risk detected	45.8%
Oil price forecast accuracy	ARIMA+XGBoost hybrid
Marine intelligence items	8 live headlines
Auto-monitoring interval	Every 5 minutes
Concurrent viewers supported	Unlimited

🔥 Auto-Monitoring — 24/7 Autonomous Operation

Aegis doesn't wait for humans. It monitors global signals every 5 minutes autonomously:

Fetches live oil price from status API
Detects anomalies automatically
Triggers full 7-agent pipeline when risk threshold exceeded
Dashboard shows live countdown timer to next scan
Zero human intervention required in production mode

🏗️ AMD Developer Cloud Deployment

# Verify MI300X is available
rocm-smi

# Check server health
curl http://localhost:8000/health

# View live marine intelligence
curl http://localhost:8000/api/marine

# Monitor logs
tail -f /opt/aegis/logs/error.log

📋 Environment Variables

Variable	Required	Description
`GROQ_API_KEY`	✅ Yes	Get free at console.groq.com
`NGROK_TOKEN`	Colab only	Get free at ngrok.com

🙏 Acknowledgements

AMD — MI300X GPU access via AMD Developer Cloud
Groq — Ultra-fast LLaMA 3.3 70B inference
lablab.ai — Hackathon platform and community
gCaptain — Live maritime news intelligence
TradeWinds — Live shipping intelligence
FastAPI — Production API framework
XGBoost — GPU-accelerated gradient boosting

📄 License

MIT License — see LICENSE for details.

Built with ❤️ for the AMD Developer Hackathon 2026

🛡️ SHIELD AGAINST CHAOS