When enriching property data using AI, the raw output isn't always perfectly aligned with your database requirements or real estate legal definitions. This guide shows you how to apply business logic "guardrails" to AI responses, ensuring that property types, lot sizes, and square footage are accurate and properly formatted before they ever reach your Monday.com board.

flowchart LR
  A["Property Address"] --> B["Gemini AI"]
  B --> C{"Code Guardrails"}
  C --> D["Monday.com Board"]

How the guardrails work

AI models (like Gemini 2.5 Flash) are great at extracting data from tax assessor or GIS databases, but they can sometimes misclassify property types based on marketing descriptions.

To fix this, we use a code step in the automation to intercept the AI's response and apply strict real estate rules based on Lot Acreage and Square Footage.

Setting up the validation flow

Follow these steps to implement the AI extraction and validation script in your automation platform (such as Zapier or Make).

1. 1

   Configure the AI Prompt

   First, prompt your AI model to return a strict JSON object. Instruct it to bypass marketing descriptions and prioritize official government records.

   {
     "sqft": [number],
     "lot_acres": [number],
     "type": "Detached/Townhome/Condo/Multi-Family/Commercial/Land",
     "neighborhood": [string]
   }

2. 2

   Add the Guardrail Script

   Add a Code step immediately after your AI action. This JavaScript snippet cleans the AI output, parses the JSON, and applies your property type rules.

   // Assume inputData contains the raw AI text and location coordinates
   const aiResponse = inputData.rawText;
   const cleanLat = inputData.googleLat;
   const cleanLng = inputData.googleLng;
   
   if (!aiResponse) return { error: "No AI data", lat: cleanLat, lng: cleanLng };
   
   try {
     // 1. Clean and parse the JSON
     let cleaned = aiResponse.replace(/```json|```/gi, "").trim();
     const start = cleaned.indexOf('{');
     let end = cleaned.lastIndexOf('}');
     const data = JSON.parse(cleaned.substring(start, end + 1));
   
     const validTypes = ["Detached", "Townhome", "Condo", "Multi-Family", "Commercial", "Land"];
     let lotSize = data.lot_acres ? parseFloat(data.lot_acres) : 0;
     let houseSize = parseInt(data.sqft) || 0;
     
     // 2. Standardize terminology
     let aiType = data.type === "Townhouse" ? "Townhome" : data.type;
     if (aiType === "Single Family") aiType = "Detached";
     
     let finalType = aiType;
   
     // 3. Apply Guardrails
     if (lotSize === 0) {
         finalType = (aiType === "Townhome") ? "Townhome" : "Condo";
     } else if (lotSize > 0 && lotSize < 0.10) {
         if (aiType === "Detached" && houseSize < 2400) {
             finalType = "Townhome";
         }
     } else if (lotSize >= 0.10) {
         finalType = aiType;
     }
   
     // 4. Fallback validation
     if (!validTypes.includes(finalType)) finalType = "Detached";
   
     // 5. Format text (Title Case for neighborhoods)
     const toTitleCase = (str) => {
       if (!str || str === "N/A") return "N/A";
       return str.toLowerCase().split(' ').map(word => 
         word.charAt(0).toUpperCase() + word.slice(1)
       ).join(' ');
     };
   
     // Return the validated payload
     output = {
       sqft: houseSize,
       type: finalType,
       neighborhood: toTitleCase(data.neighborhood),
       lot_acres: lotSize.toFixed(3),
       lat: cleanLat || null,
       lng: cleanLng || null
     };
   
   } catch (error) {
     output = { error: "Logic Error", details: error.message, lat: cleanLat, lng: cleanLng };
   }

3. 3

   Update Monday.com

   Finally, map the output variables from your Code step to your Monday.com "API Realtor Data" board. Because the data has been validated, you can safely map the type directly to your Status column and update the item to "Done".

Frequently Asked Questions

Next Steps