Data Schemas in Kotlin Notebook

After execution of a cell

val df = dataFrameOf("name", "age")(
    "Alice", 15,
    "Bob", null,
)

the following actions take place:

Columns in df are analyzed to extract data schema
Empty interface with DataSchema annotation is generated:

@DataSchema
interface DataFrameType

Extension properties for this DataSchema are generated:

val ColumnsContainer<DataFrameType>.age: DataColumn<Int?> @JvmName("DataFrameType_age") get() = this["age"] as DataColumn<Int?>
val DataRow<DataFrameType>.age: Int? @JvmName("DataFrameType_age") get() = this["age"] as Int?
val ColumnsContainer<DataFrameType>.name: DataColumn<String> @JvmName("DataFrameType_name") get() = this["name"] as DataColumn<String>
val DataRow<DataFrameType>.name: String @JvmName("DataFrameType_name") get() = this["name"] as String

Every column produces two extension properties:

Property for ColumnsContainer<DataFrameType> returns column
Property for DataRow<DataFrameType> returns cell value

df variable is typed by schema interface:

val temp = df

val df = temp.cast<DataFrameType>()

To log all these additional code executions, use cell magic

%trackExecution -all

Custom Data Schemas

You can define your own DataSchema interfaces and use them in functions and classes to represent DataFrame with a specific set of columns:

@DataSchema
interface Person {
    val name: String
    val age: Int
}

After execution of this cell in notebook or annotation processing in IDEA, extension properties for data access will be generated. Now we can use these properties to create functions for typed DataFrame:

fun DataFrame<Person>.splitName() = split { name }.by(",").into("firstName", "lastName")
fun DataFrame<Person>.adults() = filter { age > 18 }

In Kotlin Notebook these functions will work automatically for any DataFrame that matches Person schema:

val df = dataFrameOf("name", "age", "weight")(
    "Merton, Alice", 15, 60.0,
    "Marley, Bob", 20, 73.5,
)

Schema of df is compatible with Person, so auto-generated schema interface will inherit from it:

@DataSchema(isOpen = false)
interface DataFrameType : Person

val ColumnsContainer<DataFrameType>.weight: DataColumn<Double> get() = this["weight"] as DataColumn<Double>
val DataRow<DataFrameType>.weight: Double get() = this["weight"] as Double

Despite df has additional column weight, previously defined functions for DataFrame<Person> will work for it:

df.splitName()

firstName lastName age weight
   Merton    Alice  15 60.000
   Marley      Bob  20 73.125

df.adults()

name        age weight
Marley, Bob  20   73.5

Use external Data Schemas

Sometimes it is convenient to extract reusable code from Kotlin Notebook into the Kotlin JVM library. Schema interfaces should also be extracted if this code uses Custom Data Schemas.

In order to enable support them in Kotlin, you should register them in library integration class with useSchema function:

@DataSchema
interface Person {
    val name: String
    val age: Int
}

fun DataFrame<Person>.countAdults() = count { it[Person::age] > 18 }

@JupyterLibrary
internal class Integration : JupyterIntegration() {

    override fun Builder.onLoaded() {
        onLoaded {
            useSchema<Person>()
        }
    }
}

After loading this library into the notebook, schema interfaces for all DataFrame variables that match Person schema will derive from Person

val df = dataFrameOf("name", "age")(
    "Alice", 15,
    "Bob", 20,
)

Now df is assignable to DataFrame<Person> and countAdults is available:

df.countAdults()

18 September 2025