Globot

Inspiration On a Friday at 4:55 PM, the Red Sea crisis erupts. You're responsible for $50 million in cargo, drowning in Excel spreadsheets while Reuters updates flood in. Your boss is yelling. What do you do? This scenario inspired Globot — an Agentic AI system that transforms reactive crisis management into proactive defense . What it does Globot is a 5-Agent AI Reasoning Engine powered by Gemini 3 that: 🔭 Market Sentinel : Monitors geopolitical signals from Reuters/Bloomberg in real-time 🛡️ Risk Hedger : Calculates financial impact of route changes (fuel costs +$180K, freight volatility) 🚢 Logistics Orchestrator : Automatically replans shipping routes to avoid conflict zones 📋 Compliance Manager : Analyzes 500-page insurance policies using Gemini's 2M token context window ⚖️ Adversarial Debate : Red-team tests every decision to prevent AI hallucinations Key Innovation : We use Gemini Vision to analyze satellite imagery, detecting port congestion and canal blockages 6 hours before official announcements. How we built it Backend : Python + FastAPI + CrewAI (Multi-Agent Orchestration) Frontend : React + TypeScript + Deck.gl (3D Globe Visualization) AI : Gemini 3 Flash (Reasoning) + Gemini Vision (Satellite Analysis) + Gemini Embeddings (RAG) Auth : Clerk (Enterprise-grade authentication) Database : SQLite + ChromaDB (Vector Store) Challenges we ran into aiohttp Compatibility : Python 3.13 introduced breaking changes; we implemented monkey patches to maintain CrewAI compatibility Long Context Window : Feeding 500-page legal documents required careful prompt engineering Multimodal Reasoning : Getting Gemini Vision to output structured risk assessments from satellite images Accomplishments we're proud of Visual Risk Intelligence : First supply chain AI to use satellite imagery for early warning Explainable AI : Every decision shows a transparent Chain-of-Thought reasoning process Human-in-the-Loop : All critical decisions require human approval — responsible AI in action What we learned Gemini's multimodal capabilities enable use cases impossible with text-only models Long context windows unlock enterprise document understanding Multi-agent systems are more robust than single-model approaches What's next for Globot Integration with Bloomberg Terminal API for live market data MarineTraffic API for real-time AIS ship tracking Sentinel-2 Satellite API for automated port monitoring Expansion to air freight and rail logistics