CrewAI-based Credit Risk Demo with Plano Orchestration and Guardrails

demos/use_cases/credit_risk_case_copilot/.gitignore

@@ -0,0 +1,55 @@
+# Environment variables
+.env
+
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+.uv/
+
+# Virtual environments
+venv/
+ENV/
+env/
+
+# IDE
+.vscode/
+.idea/
+*.swp
+*.swo
+*~
+.DS_Store
+
+# Streamlit
+.streamlit/
+
+# Logs
+*.log
+logs/
+
+# Database
+*.db
+*.sqlite
+
+# Temporary files
+tmp/
+temp/
+*.tmp

demos/use_cases/credit_risk_case_copilot/Dockerfile

@@ -0,0 +1,28 @@
+FROM python:3.11-slim
+
+WORKDIR /app
+
+# Install system dependencies
+RUN apt-get update && \
+    apt-get install -y bash curl && \
+    rm -rf /var/lib/apt/lists/*
+
+# Install uv package manager
+RUN pip install --no-cache-dir uv
+
+# Copy dependency files
+COPY pyproject.toml README.md* ./
+COPY scenarios/ ./scenarios/
+
+# Install dependencies
+RUN uv sync --no-dev || uv pip install --system -e .
+
+# Copy application code
+COPY src/ ./src/
+
+# Set environment variables
+ENV PYTHONUNBUFFERED=1
+ENV PYTHONPATH=/app
+
+# Default command (overridden in docker-compose)
+CMD ["uv", "run", "python", "src/credit_risk_demo/risk_crew_agent.py"]

demos/use_cases/credit_risk_case_copilot/README.md

@@ -0,0 +1,73 @@
+# Credit Risk Case Copilot
+
+A small demo that follows the two-loop model: Plano is the **outer loop** (routing, guardrails, tracing), and each credit-risk step is a focused **inner-loop agent**.
+
+---
+
+## What runs
+
+- **Risk Crew Agent (10530)**: four OpenAI-compatible endpoints (intake, risk, policy, memo).
+- **PII Filter (10550)**: redacts PII and flags prompt injection.
+- **Streamlit UI (8501)**: single-call client.
+- **Jaeger (16686)**: tracing backend.
+
+---
+
+## Quick start
+
+```bash
+cp .env.example .env
+# add OPENAI_API_KEY
+docker compose up --build
+uvx planoai up config.yaml
+```
+
+Open:
+- Streamlit UI: http://localhost:8501
+- Jaeger: http://localhost:16686
+
+---
+
+## How it works
+
+1. The UI sends **one** request to Plano with the application JSON.
+2. Plano routes the request across the four agents in order:
+   intake → risk → policy → memo.
+3. Each agent returns JSON with a `step` key.
+4. The memo agent returns the final response.
+
+All model calls go through Plano’s LLM gateway, and guardrails run before any agent sees input.
+
+---
+
+## Endpoints
+
+Risk Crew Agent (10530):
+- `POST /v1/agents/intake/chat/completions`
+- `POST /v1/agents/risk/chat/completions`
+- `POST /v1/agents/policy/chat/completions`
+- `POST /v1/agents/memo/chat/completions`
+- `GET /health`
+
+PII Filter (10550):
+- `POST /v1/tools/pii_security_filter`
+- `GET /health`
+
+Plano (8001):
+- `POST /v1/chat/completions`
+
+---
+
+## UI flow
+
+1. Paste or select an application JSON.
+2. Click **Assess Risk**.
+3. Review the decision memo.
+
+---
+
+## Troubleshooting
+
+- **No response**: confirm Plano is running and ports are free (`8001`, `10530`, `10550`, `8501`).
+- **LLM gateway errors**: check `LLM_GATEWAY_ENDPOINT=http://host.docker.internal:12000/v1`.
+- **No traces**: check Jaeger and `OTLP_ENDPOINT`.

demos/use_cases/credit_risk_case_copilot/config.yaml

@@ -0,0 +1,134 @@
+version: v0.3.0
+
+# Define the standalone credit risk agents
+agents:
+  - id: loan_intake_agent
+    #url: http://localhost:10530/v1/agents/intake/chat/completions
+    url: http://host.docker.internal:10530/v1/agents/intake/chat/completions
+  - id: risk_scoring_agent
+    #url: http://localhost:10530/v1/agents/risk/chat/completions
+    url: http://host.docker.internal:10530/v1/agents/risk/chat/completions
+  - id: policy_compliance_agent
+    #url: http://localhost:10530/v1/agents/policy/chat/completions
+    url: http://host.docker.internal:10530/v1/agents/policy/chat/completions
+  - id: decision_memo_agent
+    #url: http://localhost:10530/v1/agents/memo/chat/completions
+    url: http://host.docker.internal:10530/v1/agents/memo/chat/completions
+
+# HTTP filter for PII redaction and prompt injection detection
+filters:
+  - id: pii_security_filter
+    #url: http://localhost:10550/v1/tools/pii_security_filter
+    url: http://host.docker.internal:10550/v1/tools/pii_security_filter
+    type: http
+
+# LLM providers with model routing
+model_providers:
+  - model: openai/gpt-4o
+    access_key: $OPENAI_API_KEY
+    default: true
+  - model: openai/gpt-4o-mini
+    access_key: $OPENAI_API_KEY
+
+# ToDo: Debug model aliases
+# Model aliases for semantic naming
+model_aliases:
+  risk_fast:
+    target: openai/gpt-4o-mini
+  risk_reasoning:
+    target: openai/gpt-4o
+
+# Listeners
+listeners:
+  # Agent listener for routing credit risk requests
+  - type: agent
+    name: credit_risk_service
+    port: 8001
+    router: plano_orchestrator_v1
+    address: 0.0.0.0
+    agents:
+      - id: loan_intake_agent
+        description: |
+          Loan Intake Agent - Step 1 of 4 in the credit risk pipeline. Run first.
+
+          CAPABILITIES:
+          * Normalize applicant data and calculate derived fields (e.g., DTI)
+          * Identify missing or inconsistent fields
+          * Produce structured intake JSON for downstream agents
+
+          USE CASES:
+          * "Normalize this loan application"
+          * "Extract and validate applicant data"
+
+          OUTPUT REQUIREMENTS:
+          * Return JSON with step="intake" and normalized_data/missing_fields
+          * Do not provide the final decision memo
+          * This output is used by risk_scoring_agent next
+        filter_chain:
+          - pii_security_filter
+      - id: risk_scoring_agent
+        description: |
+          Risk Scoring Agent - Step 2 of 4. Run after intake.
+
+          CAPABILITIES:
+          * Evaluate credit score, DTI, delinquencies, utilization
+          * Assign LOW/MEDIUM/HIGH risk bands with confidence
+          * Explain top 3 risk drivers with evidence
+
+          USE CASES:
+          * "Score the risk for this applicant"
+          * "Provide risk band and drivers"
+
+          OUTPUT REQUIREMENTS:
+          * Use intake output from prior assistant message
+          * Return JSON with step="risk" and risk_band/confidence_score/top_3_risk_drivers
+          * This output is used by policy_compliance_agent next
+        filter_chain:
+          - pii_security_filter
+      - id: policy_compliance_agent
+        description: |
+          Policy Compliance Agent - Step 3 of 4. Run after risk scoring.
+
+          CAPABILITIES:
+          * Verify KYC, income, and address checks
+          * Flag policy exceptions (DTI, credit score, delinquencies)
+          * Determine required documents by risk band
+
+          USE CASES:
+          * "Check policy compliance"
+          * "List required documents"
+
+          OUTPUT REQUIREMENTS:
+          * Use intake + risk outputs from prior assistant messages
+          * Return JSON with step="policy" and policy_checks/exceptions/required_documents
+          * This output is used by decision_memo_agent next
+        filter_chain:
+          - pii_security_filter
+      - id: decision_memo_agent
+        description: |
+          Decision Memo Agent - Step 4 of 4. Final response to the user.
+
+          CAPABILITIES:
+          * Create concise decision memos
+          * Recommend APPROVE/CONDITIONAL_APPROVE/REFER/REJECT
+
+          USE CASES:
+          * "Draft a decision memo"
+          * "Recommend a credit decision"
+
+          OUTPUT REQUIREMENTS:
+          * Use intake + risk + policy outputs from prior assistant messages
+          * Return JSON with step="memo", recommended_action, decision_memo
+          * Provide the user-facing memo as the final response
+        filter_chain:
+          - pii_security_filter
+
+  # Model listener for internal LLM gateway (used by agents)
+  - type: model
+    name: llm_gateway
+    address: 0.0.0.0
+    port: 12000
+
+# OpenTelemetry tracing
+tracing:
+  random_sampling: 100

demos/use_cases/credit_risk_case_copilot/docker-compose.yaml

@@ -0,0 +1,59 @@
+services:
+  # Risk Crew Agent - CrewAI-based multi-agent service
+  risk-crew-agent:
+    build:
+      context: .
+      dockerfile: Dockerfile
+    container_name: risk-crew-agent
+    restart: unless-stopped
+    ports:
+      - "10530:10530"
+    environment:
+      - OPENAI_API_KEY=${OPENAI_API_KEY}
+      - LLM_GATEWAY_ENDPOINT=http://host.docker.internal:12000/v1
+      - OTLP_ENDPOINT=http://jaeger:4318/v1/traces
+    command: ["uv", "run", "python", "src/credit_risk_demo/risk_crew_agent.py"]
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
+    depends_on:
+      - jaeger
+
+  # PII Security Filter (MCP)
+  pii-filter:
+    build:
+      context: .
+      dockerfile: Dockerfile
+    container_name: pii-filter
+    restart: unless-stopped
+    ports:
+      - "10550:10550"
+    command: ["uv", "run", "python", "src/credit_risk_demo/pii_filter.py"]
+
+  # Streamlit UI
+  streamlit-ui:
+    build:
+      context: .
+      dockerfile: Dockerfile
+    container_name: streamlit-ui
+    restart: unless-stopped
+    ports:
+      - "8501:8501"
+    environment:
+      - PLANO_ENDPOINT=http://host.docker.internal:8001/v1
+    command: ["uv", "run", "streamlit", "run", "src/credit_risk_demo/ui_streamlit.py", "--server.port=8501", "--server.address=0.0.0.0"]
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
+    depends_on:
+      - risk-crew-agent
+
+  # Jaeger for distributed tracing
+  jaeger:
+    image: jaegertracing/all-in-one:latest
+    container_name: jaeger
+    restart: unless-stopped
+    ports:
+      - "16686:16686"  # Jaeger UI
+      - "4317:4317"    # OTLP gRPC
+      - "4318:4318"    # OTLP HTTP
+    environment:
+      - COLLECTOR_OTLP_ENABLED=true

demos/use_cases/credit_risk_case_copilot/pyproject.toml

@@ -0,0 +1,29 @@
+[project]
+name = "credit-risk-case-copilot"
+version = "0.1.0"
+description = "Multi-agent Credit Risk Assessment System with Plano Orchestration"
+readme = "README.md"
+requires-python = ">=3.10"
+dependencies = [
+    "fastapi>=0.115.0",
+    "uvicorn>=0.30.0",
+    "pydantic>=2.11.7",
+    "crewai>=0.80.0",
+    "crewai-tools>=0.12.0",
+    "openai>=1.0.0",
+    "httpx>=0.24.0",
+    "streamlit>=1.40.0",
+    "opentelemetry-api>=1.20.0",
+    "opentelemetry-sdk>=1.20.0",
+    "opentelemetry-exporter-otlp>=1.20.0",
+    "opentelemetry-instrumentation-fastapi>=0.41b0",
+    "python-dotenv>=1.0.0",
+    "langchain-openai>=0.1.0",
+]
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/credit_risk_demo"]