AGENTIC AI ARCHITECTURE
Multi-Agent Swarm Architecture
A distributed, autonomous system of specialized AI agents that collaborate to detect, analyze, and predict epidemiological risks through real-time data synthesis.
System Overview
High-level architecture of the Hantavirus Early Warning System
Data Sources
X/Twitter APIReddit APITelegram GroupsFlightAwareADS-B ExchangeWHO DataCDC Reports
The Sentinel
Agent A - NLP & Social Listening
Arabic-tuned LLM processing
Symptom keyword detection
Geo-cluster analysis
Sentiment baseline tracking
The Pathfinder
Agent B - Flight & Movement
Real-time flight tracking
High-risk origin monitoring
Passenger volume analysis
Connection hub mapping
The Oracle
Lead Agent - Synthesis & Prediction
Multi-agent data fusion
Risk scoring algorithm
Predictive modeling
Escalation decisions
Risk Alerts
Geo Maps
Simulations
AI Chat
The Sentinel - Agent A
Social Listening & NLP Processing Agent
Core Responsibilities
- Continuously monitors X (Twitter), Reddit, and Telegram for symptom-related keywords in Arabic, French, and English
- Applies sentiment analysis to detect panic indicators and unusual concern patterns
- Clusters geographically-tagged posts to identify potential outbreak epicenters
- Maintains a baseline sentiment score and alerts on deviations beyond 1.5 standard deviations
Technical Stack
Python 3.11LangChainGPT-4 TurboCAMeL Tools (Arabic NLP)spaCyRedis StreamsPostgreSQL + PostGIS
Keyword Examples (Arabic)
ضيق تنفسحمىنزيفمستشفىطوارئThe Pathfinder - Agent B
Flight Tracking & Movement Analysis Agent
Core Responsibilities
- Tracks all commercial flights arriving at Beirut-Rafic Hariri International Airport (BEY)
- Scores origin airports based on rodent-borne disease prevalence and recent outbreak reports
- Correlates flight patterns with social listening spikes for temporal analysis
- Identifies high-risk connection hubs (Istanbul, Dubai, Doha) for secondary exposure paths
Technical Stack
Node.jsFlightAware APIADS-B ExchangeOpenSky NetworkTimescaleDBApache Kafka
Monitored High-Risk Origins
Santiago (SCL)Buenos Aires (EZE)Sao Paulo (GRU)Panama City (PTY)
The Oracle - Lead Agent
Data Synthesis & Predictive Intelligence Agent
Core Responsibilities
- Fuses inputs from Sentinel and Pathfinder agents using weighted ensemble scoring
- Applies epidemiological models (SEIR variants) calibrated for Hantavirus transmission dynamics
- Generates 7-day, 14-day, and 30-day risk forecasts with confidence intervals
- Autonomously triggers escalation protocols when risk exceeds predefined thresholds
Technical Stack
Python 3.11LangGraphClaude 3.5 SonnetPyMC (Bayesian)NumPyroRay (distributed)
Risk Threshold Levels
Low0-30%
Moderate31-50%
High51-75%
Critical76-100%