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

> Learn how to handle semi-structured data in Firebolt.

# Work with semi-structured data

Semi-structured data is any data that does not follow a strict tabular schema and often includes fields that are not standard SQL data types. This data typically has a nested structure and supports complex types such as arrays, maps, and structs.

Common formats of semi-structured data include:

* **JSON**— A widely used format for semi-structured data. For information on loading JSON data with Firebolt, see [Load semi-structured JSON data](/guides/loading-data/working-with-semi-structured-data/load-json-data).
* **Parquet and ORC**— Serialization formats that support nested structures and complex data types. For  information on loading Parquet data with Firebolt, see [Load semi-structured Parquet data](/guides/loading-data/working-with-semi-structured-data/load-parquet-data).

## Firebolt's approach to semi-structured data

Firebolt transforms semi-structured data using arrays and structs, enabling efficient querying. These data types allow for flexible modeling of nested and hierarchical data.

### Arrays

Firebolt supports variable-length arrays. These arrays can have arbitrary nesting levels, provided the nesting level is consistent within a column and known during table creation. For more details, see [Work with arrays](/guides/sql-dialect/arrays).

### Structs

The `STRUCT` data type allows you to group multiple attributes of varying data types into a single logical unit. This is especially useful for modeling nested or hierarchical data. For more information, see [STRUCT data type](/reference-sql/data-types/struct).

### Maps

Maps, also known as dictionaries, are not supported natively by Firebolt at the moment. However, there are
different approaches to represent them in Firebolt using arrays and structs.

* A map can be represented using two coordinated arrays—one for keys and one for values. This approach is particularly useful for JSON-like data where objects have varying keys.

* Alternatively, a map can be represented using an array of `STRUCT(key T, value U)` data type where `T` is the data type of the keys and `U` is the data type of the values. Instead of manipulating two arrays and coordinating indexes, let Firebolt do it for you and enjoy the simpler query syntax.