> ## Documentation Index
> Fetch the complete documentation index at: https://docs.buyparceldata.com/llms.txt
> Use this file to discover all available pages before exploring further.

# AI and ML workflows

> Using BPD parcel data in AI pipelines, embedding generation, and agentic workflows.

## Overview

BPD parcel data is well-suited for AI and ML applications. The structured, consistent schema across 140M parcels makes it practical for embedding generation, natural language search, and agentic property research workflows.

***

## Embedding generation

Flat file exports are the recommended delivery method for embedding workflows. Load the full parcel corpus into your pipeline and generate embeddings from parcel text fields:

```python Python theme={null}
import httpx

# Example: build a text representation of a parcel for embedding
def parcel_to_text(parcel: dict) -> str:
    parts = []
    if parcel.get("owner"):
        parts.append(f"Owner: {parcel['owner']}")
    if parcel.get("situs"):
        parts.append(f"Address: {parcel['situs']}, {parcel.get('city', '')}, {parcel.get('state', '')}")
    if parcel.get("acres"):
        parts.append(f"Acres: {parcel['acres']}")
    if parcel.get("land_use"):
        parts.append(f"Land use: {parcel['land_use']}")
    if parcel.get("cdl_majority_category"):
        parts.append(f"Crop cover: {parcel['cdl_majority_category']} ({parcel.get('cdl_majority_percent', '')}%)")
    if parcel.get("zoning"):
        parts.append(f"Zoning: {parcel['zoning']}")
    if parcel.get("parcel_value"):
        parts.append(f"Assessed value: ${parcel['parcel_value']:,.0f}")
    return " | ".join(parts)
```

This approach works with any embedding model. For large-scale batch embedding, platforms like Databricks are commonly used to parallelize embedding generation across the full dataset.

***

## Natural language search

Once embeddings are generated, you can implement natural language property search:

```python Python theme={null}
# Query example: "large agricultural parcels in Iowa with corn cover"
# 1. Embed the query text
# 2. Retrieve semantically similar parcels from your vector index
# 3. Optionally post-filter with BPD API structured filters

API_KEY = "YOUR_API_KEY"
BASE = "https://api.buyparceldata.com"

# Structured query complement — filter by field after vector retrieval
response = httpx.post(
    f"{BASE}/parcels/query",
    headers={"Authorization": f"Bearer {API_KEY}"},
    json={
        "filters": [
            {"field": "state_fp", "op": "eq", "value": "19"},
            {"field": "acres", "op": "gte", "value": 100},
            {"field": "cdl_majority_category", "op": "ilike", "value": "Corn"},
        ],
        "order": {"field": "acres", "direction": "desc"},
        "limit": 50,
    },
)
parcels = response.json()["results"]
```

***

## Agentic workflows

BPD integrates cleanly into agentic AI workflows that need to answer questions about property, ownership, or land use. Typical patterns:

* **Geographic research agent**: accepts a location or address, calls `/parcels/point` or `/parcels/area`, returns structured parcel data to the agent context
* **Owner lookup agent**: queries `/parcels/query` with `owner` + `state_fp` filters to find all parcels owned by a given entity
* **Land screening agent**: applies multi-field filters (acreage, crop type, zoning, adjacency) to identify candidate parcels for a specific use case

The API's structured filter system maps naturally to agent tool parameters. Each filter field and operator can be exposed as a typed tool argument.

***

## Flat files vs. API for AI use cases

| Use case                                 | Recommended |
| ---------------------------------------- | ----------- |
| Batch embedding generation               | Flat files  |
| Building a vector index over all parcels | Flat files  |
| Real-time parcel lookup in an agent      | API         |
| Filtering parcels by structured criteria | API         |
| Offline analysis and model training      | Flat files  |

See [Flat files](/guides/flat-files) for bulk delivery options.