An enterprise-grade, Multi-Agent intelligence for high-fidelity regional economic analysis, labor market evaluation, and cross-industry site selection. Upgraded to Vertex AI Agent Engine (ADK 2.0 / AdkApp) with a 100% Live-API grounded architecture and Zero LangChain/LangGraph dependencies for massive speed-up.

The Economic Research Agent (ERA) is a production-grade site-selection and market intelligence consultant. Designed for corporate research, ERA automates comparative metropolitan cost-modeling by orchestrating live public APIs (FRED, BLS, CENSUS, HUD, EIA) into McKinsey-style strategic briefs—now upgraded with an automated Auditor Judge Agent (via Serper Live Search) for zero-hallucination verification.

User: "Compare Austin and Raleigh for a new data center HQ."

ERA: "I am analyzing technical and fiscal ROI for NAICS 518210 (Data Processing).

1. Utility Benchmarking: Texas (ERCOT) industrial rates average 8.2c/kWh, while North Carolina (Duke Energy) is 7.5c/kWh.
2. Talent Pipeline: UT Austin graduated 1,200 CS majors in 2023 vs. 850 from NC State.
3. Fiscal Climate: NC has a scheduled corporate tax phasedown to 0% by 2030, offering a better longitudinal OPEX profile. Recommendation: Raleigh for long-term tax stability; Austin for immediate talent density."

• Live Wage Analysis: Real-time median hourly wages fetched via live FRED search (No hardcoded mocks).
• Unemployment Trends: 10-year historical time-series sampling for MSA-level analysis.
• Union Density: Live state-level union membership percentages.

• Energy Matrix: Live Industrial electricity rates (per kWh) using compliant EIA IND sector codes.
• ROI Modeling: Real estate acquisition ROI based on live macro health indicators.

• Campaign Finance: Correlate political stability with corporate and PAC contribution data.
• Lobbying Hubs: Identification of industry influence and regulatory engagement levels.
• Regulatory Monitoring: Live notices from the Federal Register regarding industry-specific policy shifts.

• Workforce Burden Analysis: Correlation of Fair Market Rents (FMR) against Area Median Income (AMI).
• Relocation COLA: Precise cost-of-living benchmarking for talent retention strategy.
• Demographic Depth: Hyper-localized education and age-bucket analysis (Census ACS).

graph TD
    User([User Query]) --> Planner["Researcher Agent (Planner)"]
    
    subgraph "Structured Live Grounding (ReAct)"
        Planner --> ToolRouter{"Tool Router"}
        ToolRouter --> Macro["Macro Hub (FRED, BEA, Census, Tax Foundation)"]
        ToolRouter --> Labor["Labor Matrix (BLS, Talent Pipeline)"]
        ToolRouter --> Policy["Policy & Volatility (FEC, Regulatory, Political)"]
        ToolRouter --> Infra["Infrastructure & Climate (EIA, HUD, Resilience)"]
        ToolRouter --> Specialized["Specialized Synthesis (Metro Matrix, Relocation, Trade)"]
    end
    
    Macro --> LiveAPIs([Structured Public APIs])
    Labor --> LiveAPIs
    Policy --> LiveAPIs
    Infra --> LiveAPIs
    Specialized --> LiveAPIs
    
    LiveAPIs -->|"Grounded Data"| Planner
    
    Planner --> Judge["Auditor Judge Agent (Critic)"]
    Judge --> Search["Serper.dev Live Search"]
    Search -->|"Context Tracking"| Judge
    
    Judge --> Narrative["Narrative Synthesis & Scribe"]
    Narrative -->|"[A2UI] Response"| User
    
    style Planner fill:#f9f,stroke:#333,stroke-width:2px
    style Judge fill:#ffcc99,stroke:#333,stroke-width:2px
    style Search fill:#bbf,stroke:#333,stroke-width:2px

The ERA uses a modular grounding strategy. Set these in your .env file (see .env.example).

ERA uses uv for lightning-fast dependency management.

# Create and synchronize the virtual environment
uv sync --dev

Before deploying to the Vertex AI Reasoning Engine, ensure your local environment is authenticated with Google Cloud:

1. Install the Google Cloud CLI: Follow the installation guide.
2. Set your active project:

   gcloud config set project YOUR_PROJECT_ID

3. Authenticate your credentials:

   gcloud auth application-default login

We highly recommend setting up and deploying this agent using the Agent Starter Pack (ASP). ASP provides a production-ready framework that includes:

• Automated CI/CD Pipelines: Pre-configured GitHub Actions for streamlined Vertex AI deployments.
• Standardized Structure: Adheres to Google Cloud best practices for modular agent repositories.
• Interactive CLI Setup: Guides you through provisioning staging buckets and Vertex AI resources automatically.

# Create and activate a virtual environment
python -m venv .venv && source .venv/bin/activate

# Install the starter pack and create your project
pip install --upgrade agent-starter-pack
agent-starter-pack create my-economic-research-agent -a adk@economic-research-agent

uvx agent-starter-pack create my-economic-research-agent -a adk@economic-research-agent

ERA offers multiple interaction protocols:

# 🧠 Option 1: Interactive CLI Session (Standard)
make run

# 🛰️ Option 2: Multi-Protocol MCP Server (For Claude/Cursor)
make mcp

The ERA is designed for modular growth:

• Modifying the Persona: Edit economic_research/prompt.py to change the consultative tone.
• Adding New Skills: Add your skill in economic_research/tools/, then register it in economic_research/agent.py.
• Altering Data Flows: Use the shared_libraries/helper.py to add new HTTP/JSON normalization patterns for regional data.

How do we know ERA is accurate?

• Golden Suite: We use a 21-question integration suite (tests/integration/) targeting specific NAICS scenarios.
• Grounding Fidelity Metric: The eval/run_eval.py script uses LLM-as-a-Judge (Gemini 3.1 Pro) to verify if the output contains actual numerical data from the APIs.
• Regression Testing: pytest handles unit-level verification of API response parsing.

# Run the full 21-question validation suite
uv run pytest tests/integration/test_full_golden_suite.py

The ERA is built for the Vertex AI Reasoning Engine (ADK 2.0).

# 🌍 Step 1: Deploy to Google Cloud (Reasoning Engine)
make deploy

The ERA is engineered for Enterprise Privacy within the Google Cloud perimeter:

• Zero Data Retention: No local databases or static tables are used. Data is processed in-memory.
• Key-Safe Architecture: Secrets are managed via .env or Google Secret Manager.

Built for the Atomic Agents Initiative.