Autonomous agent cross-referencing public records to detect childcare provider fraud via building data, licensing, and imagery.
## 🔍 AI Fraud Investigation Agent An AI-powered autonomous fraud investigation agent that cross-references childcare provider licensing records against physical building data to detect anomalies. The agent uses public data — Cook County property records, Illinois DCFS licensing, Google Maps, and the Secretary of State — to find facilities where the physical evidence doesn't match the paperwork. ### Features - Searches Illinois DCFS licensing database for providers by ZIP code - Cross-references licensed capacity against actual building square footage from Cook County GIS records - Applies IL building code math to calculate maximum legal childcare occupancy - Analyzes Google Street View imagery to verify a facility looks like a real childcare center - Checks Google Places for business status, rating, and whether the address shows a different business entirely - Verifies business entity registration with the Illinois Secretary of State - Discovers cross-provider patterns: shared owners, address clusters, entities with no public footprint - Narrates its full investigation in real time — reasoning is visible as the agent works ### How to get Started? 1. Clone the GitHub repository ```bash git clone https://github.com/Shubhamsaboo/awesome-llm-apps.git cd advanced_ai_agents/single_agent_apps/ai_fraud_investigation_agent ``` 2. Install the required dependencies ```bash pip install -r requirements.txt ``` 3. Get your OpenRouter API Key - Sign up at [openrouter.ai](https://openrouter.ai) and create an API key - The free tier is sufficient for demo investigations - Default model: `anthropic/claude-sonnet-4.6` 4. Get your Google Maps API Key *(optional — skips visual analysis if omitted)* - Create a project at [console.cloud.google.com](https://console.cloud.google.com) - Enable: **Geocoding API**, **Places API**, **Street View Static API** - Create an API key and restrict it to these three APIs 5. Run the Streamlit App ```bash streamlit run fraud_investigation_agent.py ``` ### How it Works? The AI Fraud Investigation Agent uses 7 specialized tools, all powered by public data sources: - **Provider Search** — Queries the Illinois DCFS licensing portal for all active providers in a ZIP code, returning capacity, license type, and license status - **Property Analysis** — Pulls building square footage, lot size, property class, and year built from the Cook County Assessor's open data (Socrata API, no auth required) - **Capacity Calculation** — Applies Illinois DCFS Part 407 building code math: `(building_sqft × 0.65) ÷ 35 = max legal children`. A 900 sq ft building cannot legally serve 50 children — this is a mathematical impossibility, not an opinion - **Street View** — Captures four-directional Google Street View images to check whether the address looks like a real childcare facility or something else entirely - **Places Info** — Queries Google Places for the current business listed at the address, its operating status, rating, and recent reviews - **Business Registration** — Probes the Illinois Secretary of State to verify the provider is a registered legal entity - **Geocoding** — Converts addresses to coordinates for spatial analysis The agent investigates each provider in the ZIP code, narrating its reasoning as it works. When it notices something suspicious — a building too small for its license, a closed storefront claiming to run childcare, a name appearing across multiple providers — it follows that thread and explains why it matters. ### What the Agent Can (and Cannot) Detect **Can detect:** - Licensed capacity physically impossible for the building size - Addresses where Google shows a different business or a closed/vacant building - Providers with no Google listing, no reviews, no business registration - Shared owner names or agents appearing across multiple providers **Cannot detect:** - Attendance fraud (billing for children who didn't show up) — requires non-public CCAP billing records - Any fraud requiring access to internal DHS or county billing data All findings are investigative leads, not legal conclusions. The agent uses language like "requires further investigation" and "exhibits anomalies" — never "fraud." ### Geographic Scope This demo covers **Cook County, Illinois only**. Property data is sourced from the Cook County Assessor's open data, so the ZIP code selector is limited to 10 high-density Chicago neighborhoods known for concentrations of subsidized childcare providers: | ZIP | Neighborhood | |-----|-------------| | 60623 | Little Village / North Lawndale | | 60629 | Chicago Lawn | | 60644 | Austin | | 60621 | Englewood | | 60628 | Roseland | | 60619 | Chatham / Auburn Gresham | | 60636 | West Englewood | | 60612 | Near West Side | | 60620 | Auburn Gresham | | 60624 | Garfield Park | The full [Surelock Homes](https://github.com/oso95/Surelock-Homes) system supports Minnesota and additional Illinois counties with a FastAPI backend, streaming dashboard, and offline mode.