Macro RAG: Multi-Agent G20 Macroeconomic Intelligence & Copilot

An enterprise-grade, full-stack macroeconomic analysis platform, risk simulation lab, and AI copilot designed for G20 nations. The application integrates automated data ingestion from official APIs, multi-layered quality validation, dynamic simulation, and generative AI (Gemini API) to deliver real-time economic insights and reporting.

---

🚀 Key Features

• Executive Dashboard: Real-time G20 indicator tracking (GDP, CPI, Unemployment) with dynamic geospatial world heatmaps, US Treasury yield curves, and interactive alert watchlist configurations.
• Macro Copilot: Advanced chat assistant powered by Gemini 2.5 Flash with dense (ChromaDB) and sparse (BM25) search fusion, summaries, detailed analyses, and live indicators attribution.
• Enterprise Confidence Engine: Every AI-generated response includes a dynamic, backend-calculated confidence score (High/Medium/Low) based on RAG retrieval scores, citation coverage, source credibility, and LLM logical checks, supporting graceful weighting adaptation for RAG and non-RAG pipelines.
• AI Quality Center: An executive evaluation dashboard displaying RAGAS and TruLens benchmark metrics, datasets parameters, and historical evaluation runs to transparently showcase system reliability.
• RAGAS Benchmark Evaluation: Built-in support for measuring RAG quality using metrics like Faithfulness, Answer Relevancy, Context Precision, and Context Recall.
• TruLens Triad Evaluation: Measures pipeline fidelity through Groundedness, Answer Relevance, and Context Relevance checks, alongside token costs and latency metrics.
• Risk Simulation Lab: Adjust economic levers (GDP growth, policy interest rates, inflation) to generate probabilistic economic scenarios (stagflation, overheating, recession) with structural trade-offs.
• Custom Report Generator: Auto-drafts publication-ready reviews with CSS-styled charts, data summaries, and bibliographic records.
• Multi-Layered Validation Engine: Multi-weighted data checks (Source, Format, Quality, and Range) to automatically approve clean data or route anomalies to an admin review queue.
• Human-in-the-loop Validation: Dedicated administrative dashboard displaying flagged anomalous data synchronization payloads with full approve/reject capability.
• LangSmith Observability: Complete nesting tracing for retrieval-to-generation chains, operating with production-safe fallback when tracing is disabled.

---

🏗️ Updated AI Architecture Flow

The evaluation, validation, and observability services form a first-class loop inside the G20 intelligence platform:

    User Query
        │
        ▼
  Question Router
        │
        ├──► [LIVE_DATA / BOTH] ───────► FRED SQLite KPI Retrieval
        │
        └──► [DOCUMENT_RETRIEVAL / BOTH]
                     │
                     ▼