CustomSchema.java

/*
 * Custom fields on top of a preset, fields injected into every line-item row, and saving
 * the combination so later calls are one argument.
 *
 *   java -cp target/visionapi-java-1.0.0.jar examples/CustomSchema.java invoice.pdf
 */

import io.visionapi.AnalyzeRequest;
import io.visionapi.AnalyzeResponse;
import io.visionapi.ConflictException;
import io.visionapi.Row;
import io.visionapi.Schema;
import io.visionapi.VisionApi;
import java.nio.file.Path;
import java.util.Map;

public class CustomSchema {

    public static void main(String[] args) {
        Path path = Path.of(args.length > 0 ? args[0] : "invoice.pdf");
        VisionApi vision = VisionApi.create();

        Map<String, Object> schema = Schema.builder()
                // Plain form: the string is the description. Descriptions are the prompt —
                // the more precisely you say what you want, the better the extraction.
                .field("machine_serial", "Serial number of the machine being invoiced, without the \"SN:\" prefix")
                // Typed forms. Numbers come back as JSON numbers, dates as ISO YYYY-MM-DD.
                .number("total_net", "Total before tax")
                .bool("is_paid", "Whether the invoice is stamped or marked PAID")
                .array("reference_numbers", "All reference numbers", "string")
                // lineItem is reserved: these become extra columns on every row of the
                // preset's line-item array. Only meaningful with a preset that has line
                // items; with a bare custom schema there is nothing to inject into.
                .lineItem(Schema.of("lot_number", "The lot number printed on this line, if present"))
                .build();

        AnalyzeResponse res = vision.analyze(AnalyzeRequest.builder()
                .file(path)
                .preset("invoice")
                .schema(schema)
                .build());

        System.out.println("serial: " + res.result().field("machine_serial").asString().orElse("—"));
        System.out.println("net total: " + res.result().field("total_net").asDouble().orElse(0));
        System.out.println("paid: " + res.result().field("is_paid").asBoolean().orElse(false));

        for (Row row : res.result().rows("line_item")) {
            System.out.printf("  %s × %s = %s  [lot %s]%n",
                    row.field("quantity").value(),
                    row.field("description").asString().orElse("(no description)"),
                    row.field("amount").value(),
                    row.field("lot_number").asString().orElse("—"));
        }

        // Save it, and the next call is schemaName("our-invoices"). The definition is
        // compiled before it is stored, so an invalid schema fails here rather than on the
        // first extraction that uses it.
        try {
            vision.createSchema("our-invoices", "invoice", schema);
            System.out.println("\nsaved as \"our-invoices\"");
        } catch (ConflictException e) {
            vision.updateSchema("our-invoices", "invoice", schema);
            System.out.println("\nupdated \"our-invoices\"");
        }

        AnalyzeResponse again = vision.analyze(AnalyzeRequest.builder()
                .file(path)
                .schemaName("our-invoices")
                .build());
        System.out.println("same fields via the saved schema: " + again.result().names().size());
    }
}