AEGIS - Agentic Global Intelligence System

Real-time commodity stress monitoring and AI-powered hedging strategy generation, built on a multi-agent architecture powered by Amazon Nova Pro.

Comment

Inspiration

AEGIS was inspired by a practical problem observed in financial markets: large commodity price movements are often triggered by breaking geopolitical news, but synthesizing those signals quickly is difficult. Analysts typically need to read research, examine price data, and assess risk before deciding on a hedge. This process can take hours.

At the same time, recent advances in AI systems have enabled the design of multi-agent architectures, in which specialized AI components analyze different sources of information and combine their outputs.

AEGIS was built to explore how a multi-agent AI pipeline could automate this signal synthesis by combining market data and news analysis into a single risk indicator.

What it does

AEGIS is a multi-agent AI system that monitors global commodity markets and geopolitical news in real time, synthesizes both signals into a single Global Stress Index, and generates hedging strategies for a user's specific portfolio, all powered by Amazon Nova Pro on AWS Bedrock. Three specialized agents run in parallel:

  • Commodity Agent - fetches 90 days of live price data across Oil, Gold, Wheat, Copper, and Natural Gas, and asks Amazon Nova Pro to reason about volatility regimes, price deviations, and cross-commodity correlations
  • News Agent - reads live headlines from major financial news sources and institutional research, and asks Nova Pro to classify each one into geopolitical signal buckets: war and conflict, sanctions, supply disruptions, policy shifts, etc.
  • Hedge Agent - receives the Global Stress Index plus the user's portfolio and asks Nova Pro to generate three specific, instrument-level hedging strategies with urgency ratings and rationale

The Orchestrator blends the agent scores into the Global Stress Index using sigmoid normalization, ensuring that the inputs are mathematically commensurable before averaging. Results are surfaced through a Streamlit dashboard, with a full agent reasoning trace; every step Nova took is visible and auditable.

How we built it

AEGIS is built on a fully agentic architecture powered entirely by Amazon Nova Pro via AWS Bedrock:

  • Amazon Bedrock - hosts all three Nova Pro agent calls, handling commodity scoring, news classification, and hedging strategy generation
  • Amazon Nova Pro - the reasoning engine across all agents, chosen for its ability to handle complex multi-step reasoning across independent contexts
  • yfinance - live 90-day commodity price history
  • NewsAPI + institutional sources - live headlines from trusted financial institutions
  • Python - agent orchestration and pipeline logic
  • Streamlit - front-end dashboard with real-time results and reasoning trace; app deployment

Challenges we ran into

Limited domain knowledge

Building the system required learning basic concepts in commodity markets and risk management to ensure outputs were reasonable and interpretable.

Combining different types of signals

Commodity stress metrics are naturally bounded, while geopolitical signals from news are not. To combine them, the system applies sigmoid normalization, mapping scores to a consistent scale before aggregation.

Designing a reliable multi-agent workflow

Ensuring that each agent produced structured outputs and that results could be combined consistently required several iterations in prompt design and orchestration logic.

Accomplishments that we're proud of

  • Built a working multi-agent AI pipeline where three specialized agents analyze different data sources and produce results in seconds
  • Designed a Global Stress Index that combines market data and news signals into a single interpretable metric
  • Implemented a transparent reasoning trace so users can inspect how each recommendation was generated
  • Created a modular architecture that allows new agents or data sources to be added easily

What we learned

Multi-agent architectures can help organize complex reasoning tasks. Separating market analysis, news analysis, and strategy generation improved clarity and modularity. Transparency is important when AI is used in decision-support systems. Providing a reasoning trace helps users understand and evaluate model outputs. Domain understanding improves system quality. Even when using AI models, a basic financial context was necessary to structure the pipeline effectively.

What's next for AEGIS - Agentic Global Intelligence System

Potential next steps include:

  • Expanding commodity coverage to include additional energy, metals, and agricultural markets for broader risk visibility.
  • Validate hedge recommendations using historical market data and past volatility events.
  • Integrating quantitative hedging frameworks commonly used in finance to strengthen the strategies generated by the hedge agent.
  • Real-time alerts when the Global Stress Index crosses predefined thresholds or when high-impact geopolitical signals are detected.
  • Additional instrument support, including options and swaps, to allow more flexible hedging approaches.

Built With

Share this project:

Updates