Financial Services Application

This is a financial services application that evaluates customer eligibility for financial products.

---

System Architecture

Our solution is built using Google ADK with two cooperating agents:

• Profile Agent (Agent A): retrieves and normalizes customer profiles via MCP.
• Eligibility Agent (Agent B): evaluates product eligibility using a Qdrant-based RAG retriever and outputs a structured decision.

These are orchestrated by a root Sequential Agent that validates inputs and coordinates the workflow end-to-end.

The codebase follows a modular structure:

• Agents
• Schema models
• Qdrant retriever
• Config
• Toolsets

This ensures clarity, maintainability, and scalability.

---

Key Features

• Explainability & Traceability: Uses Ariza AI. Each run generates a full JSON trace including profile snapshot, retrieved context, tool calls, policy checks, and final decision with confidence.

• Robustness: Gracefully handles missing fields, low credit scores, and regulatory restrictions (restricted countries).

• Security: All sensitive credentials are stored in environment variables.

• Performance: End-to-end decision latency < 20s (due to free plan usage of Google AI Studio, Qdrant, and Postgres via Docker).

---

Testing

Deterministic unit tests confirm correct handling of:

• Missing profile fields
• Low credit score
• Regulatory restriction

---

Running the Application

Using adk web

1. Activate your virtual environment:

   uv init .
   uv sync

2. Run toolbox:

   ./toolbox --tools-file "tools.yaml" --address 0.0.0.0 --port 5000

3. Run Qdrant and Postgres using Docker.

4. Copy environment variables file:

   cp .env.example .env

5. Start the ADK web server:

   adk web

6. Open the web interface: http://127.0.0.1:8000

---