Aegis

Summary

Aegis is an open-source applied economics intelligence platform organized around 18 analytical layers and 257 modules. It combines a structured estimation engine, named data collectors, and an AI brain that composes evidence into briefings traceable to source. The platform supports a five-layer government digital twin methodology used to translate ministry questions into reproducible analysis.

What it is

Aegis unifies estimation, ingestion, and briefing generation into a single FastAPI application that economists and policy teams operate from a CLI, an HTTP API, and a Next.js front end. The 18 analytical layers cover trade, macro, labor, development, agricultural, integration, financial, health, environmental, public, spatial, political, behavioral, industrial, monetary, energy, demographic, and methods. Each module follows a strict LayerBase contract that returns numerical results with a citation field, and all signals are classified on a four-level scale of stable, watch, stress, and crisis.

The estimation engine ports twelve methods including OLS, IV, panel fixed effects, difference in differences, regression discontinuity, double machine learning, causal forest, synthetic DiD, staggered DiD, shift share, bounds, and randomization inference. Thirteen collectors pull from FRED, World Bank WDI, ILO, FAOSTAT, BLS, IMF WEO, Penn World Table, UN Comtrade, USDA, NOAA, V-Dem, PovcalNet, and WHO using a common collect, validate, and store pipeline. The AI brain runs Claude Sonnet with 24 structured tools across a ten-round agentic loop, producing economic conditions, trade flash, and country deep dive briefings. A five-layer government digital twin methodology maps a ministry mandate to data sources, modules, integration scores, and a final briefing template, giving policy teams a reproducible path from question to evidence.

Methodology

• Eighteen analytical layers spanning trade, macro, labor, development, agricultural, integration, financial, health, environmental, public, spatial, political, behavioral, industrial, monetary, energy, demographic, and methods.
• 257 modules built on a single LayerBase contract with signal classification across stable, watch, stress, and crisis bands.
• Five-layer government digital twin methodology that maps a ministry mandate to data, modules, integration, and briefing.
• Three production briefing templates: Economic Conditions, Trade Flash, and Country Deep Dive, each composed from layer scores.
• Estimation engine with twelve methods including OLS, IV, panel FE, DiD, RDD, double ML, causal forest, synthetic DiD, staggered DiD, shift share, bounds, and randomization inference.
• Integration layer that combines composite scoring, attribution, spillover, cross correlation, and scenario simulation.
• AI brain with 24 structured tools and a ten-round agentic loop that grounds every claim in tool output and citations.
• Knowledge base compiled from analysis results with fact extraction, article compilation, and a 30-day staleness sweep.

Data sources

• FRED (Federal Reserve Economic Data)
• World Bank WDI
• IMF World Economic Outlook
• UN Comtrade
• ILO and BLS labor statistics
• FAOSTAT and USDA agricultural data
• Penn World Table, PovcalNet, V-Dem, WHO
• NOAA climate and weather data

Deliverables when used in engagements

• Country and regional economic conditions briefings with composite layer scores.
• Trade flash assessments covering tariffs, sanctions, and bilateral exposure.
• Country deep dive reports with citations to primary data sources.
• Reproducible estimation outputs from twelve econometric methods with figures and tables.
• AI assisted analyses delivered through CLI, HTTP API, and a Next.js front end.