This workflow combines address formatting, Google Maps geocoding, and Gemini AI to automatically extract property details and generate map URLs. By chaining these tools together, you can take raw, messy address inputs and transform them into standardized, validated property records complete with satellite and street-view imagery.

flowchart TD
    A["Raw Address & Client Input"] --> B["Format & Clean Address"]
    B --> C["Google Maps Geocoding"]
    C --> D["Gemini AI Extraction"]
    D --> E["Parse & Validate JSON"]
    E --> F["Generate Map URLs & Final Output"]

How the workflow operates

The process is broken down into four main stages, ensuring that data is cleaned before it is searched, and validated before it is saved.

1. 1

   Format Address and Client Name

   First, the workflow takes raw input fields (Street, City, State, Zip, Unit, and Client Name) and standardizes them. It applies Title Case to names and cities, cleans up unit labels, and constructs a pristine fullAddress string. This ensures the geocoding step has the highest possible match rate.

2. 2

   Geocode the Location

   Using the cleaned fullAddress, the workflow calls the Google Maps Geocoding API. This translates the text address into precise latitude and longitude coordinates (navLat and navLng), which are essential for pulling accurate map images later.

3. 3

   Extract Property Specs with AI

   Next, Gemini 2.5 Flash is prompted to act as a high-speed property data extractor. It uses the formatted address and coordinates to search accessible real estate databases (like Zillow, Redfin, or local GIS records).

   The AI is instructed to return a strict JSON object containing:

   • Square footage (sqft)

   • Lot acreage (lot_acres)

   • Property type (ai_type)

   • Neighborhood

4. 4

   Analyze Data and Generate URLs

   Finally, a code step parses the AI's JSON response. It compares the AI-detected property type against the client's requested property type to flag any mismatches. It also uses the coordinates to generate direct URLs for Google Street View and Satellite imagery.

Ensuring reliable AI JSON parsing

When asking LLMs to return JSON, they occasionally include markdown formatting (like json ) or conversational preamble. This workflow uses a robust extraction method to guarantee the data parses correctly.

The AI prompt instructs Gemini to wrap the final JSON object in custom @@@ markers:

@@@
{
  "sqft": 2450,
  "lot_acres": 0.25,
  "ai_type": "Detached",
  "neighborhood": "Oakwood Estates"
}
@@@

The validation step then uses regular expressions to extract exactly what is between those markers, stripping out any unexpected text. If the markers fail, it falls back to finding the first valid { } block in the response.

Validation rules and edge cases

The final step of the workflow evaluates the extracted data and assigns a status_icon and validation_status based on specific business rules.

Expected Output

Once the workflow completes, it returns a structured payload ready to be sent to your database or CRM.