arsvendg/Stirling-PDF
1
0
Fork 0
mirror of https://github.com/arsvendg/Stirling-PDF.git synced 2026-07-17 11:45:05 +02:00
Code Issues Actions 3 Packages Projects Releases 1 Wiki Activity

Feature/pdf to markdown agent (#6271)

Browse Source 
Co-authored-by: James Brunton <[email protected]>
This commit is contained in:
EthanHealy01
2026-05-14 16:20:45 +00:00
committed by GitHub
co-authored by James Brunton
parent 5b9ef852ab
commit ece1bb6865
36 changed files with 4081 additions and 267 deletions
Download Patch File Download Diff File Expand all files Collapse all files
app/proprietary/src/main/java/stirling/software/proprietary/model/api/ai/AiWorkflowOutcome.java
+2 -1
View File
@@ -20,7 +20,8 @@ public enum AiWorkflowOutcome {
TOOL_CALL("tool_call"),
COMPLETED("completed"),
UNSUPPORTED_CAPABILITY("unsupported_capability"),
CANNOT_CONTINUE("cannot_continue");
CANNOT_CONTINUE("cannot_continue"),
GENERATE_FILE("generate_file");
private final String value;
app/proprietary/src/main/java/stirling/software/proprietary/model/api/ai/AiWorkflowResponse.java
+10
View File
@@ -4,6 +4,8 @@ import java.util.ArrayList;
import java.util.List;
import java.util.Map;
import com.fasterxml.jackson.annotation.JsonProperty;
import io.swagger.v3.oas.annotations.media.Schema;
import lombok.Data;
@@ -20,6 +22,14 @@ public class AiWorkflowResponse {
@Schema(description = "Answer returned by the AI workflow when applicable")
private String answer;
@JsonProperty("content")
@Schema(description = "Text content to package as a file (generate_file outcomes)")
private String generatedContent;
@JsonProperty("filename")
@Schema(description = "Desired output filename for generate_file outcomes")
private String generatedFilename;
@Schema(description = "Summary returned by the AI workflow when applicable")
private String summary;
app/proprietary/src/main/java/stirling/software/proprietary/pdf/FlexibleCSVWriter.java
-17
View File
@@ -1,17 +0,0 @@
package stirling.software.proprietary.pdf;
import org.apache.commons.csv.CSVFormat;
import technology.tabula.writers.CSVWriter;
/** Exposes Tabula's protected {@link CSVWriter#CSVWriter(CSVFormat)} constructor. */
public class FlexibleCSVWriter extends CSVWriter {
public FlexibleCSVWriter() {
super();
}
public FlexibleCSVWriter(CSVFormat csvFormat) {
super(csvFormat);
}
}
app/proprietary/src/main/java/stirling/software/proprietary/service/AiWorkflowService.java
+24
View File
@@ -160,6 +160,7 @@ public class AiWorkflowService {
case TOOL_CALL -> onToolCall(response, filesById, listener);
case PLAN -> onPlan(response, filesById, request, listener);
case ANSWER -> onAnswer(response, filesById, request, listener);
case GENERATE_FILE -> onGenerateFile(response, listener);
case NOT_FOUND,
NEED_CLARIFICATION,
CANNOT_DO,
@@ -364,6 +365,29 @@ public class AiWorkflowService {
return new WorkflowState.Terminal(response);
}
private WorkflowState onGenerateFile(AiWorkflowResponse response, ProgressListener listener)
throws IOException {
String content = response.getGeneratedContent();
String filename = response.getGeneratedFilename();
if (content == null || filename == null || filename.isBlank()) {
return new WorkflowState.Terminal(
cannotContinue(
"AI engine returned generate_file without content or filename."));
}
listener.onProgress(AiWorkflowProgressEvent.of(AiWorkflowPhase.PROCESSING));
String safeFilename = Filenames.toSimpleFileName(filename);
byte[] bytes = content.getBytes(java.nio.charset.StandardCharsets.UTF_8);
org.springframework.core.io.Resource resource =
new org.springframework.core.io.ByteArrayResource(bytes) {
@Override
public String getFilename() {
return safeFilename;
}
};
return new WorkflowState.Terminal(
buildCompletedResponse(response.getSummary(), List.of(resource), List.of(), null));
}
@SuppressWarnings("unchecked")
private WorkflowState runPlan(
List<Map<String, Object>> steps,
app/proprietary/src/main/java/stirling/software/proprietary/service/PdfContentExtractor.java
+95 -20
View File
@@ -3,7 +3,6 @@ package stirling.software.proprietary.service;
import java.io.IOException;
import java.io.StringWriter;
import java.util.ArrayList;
import java.util.Collections;
import java.util.LinkedHashSet;
import java.util.List;
import java.util.Map;
@@ -12,6 +11,7 @@ import java.util.Set;
import java.util.stream.Collectors;
import org.apache.commons.csv.CSVFormat;
import org.apache.commons.csv.CSVPrinter;
import org.apache.commons.csv.QuoteMode;
import org.apache.pdfbox.pdmodel.PDDocument;
import org.apache.pdfbox.text.PDFTextStripper;
@@ -21,25 +21,30 @@ import com.fasterxml.jackson.annotation.JsonIgnore;
import com.fasterxml.jackson.annotation.JsonValue;
import lombok.Data;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
import stirling.software.SPDF.pdf.parser.PdfIngester;
import stirling.software.SPDF.pdf.parser.PdfModels.ParsedPage;
import stirling.software.SPDF.pdf.parser.PdfModels.RawLine;
import stirling.software.SPDF.pdf.parser.PdfModels.TableFragment;
import stirling.software.SPDF.pdf.parser.PdfModels.TextFragment;
import stirling.software.SPDF.pdf.parser.TabulaTableParser;
import stirling.software.common.util.ExceptionUtils;
import stirling.software.common.util.PdfUtils;
import stirling.software.proprietary.model.api.ai.AiPdfContentType;
import stirling.software.proprietary.model.api.ai.AiWorkflowFileRequest;
import stirling.software.proprietary.model.api.ai.AiWorkflowTextSelection;
import stirling.software.proprietary.model.api.ai.FolioType;
import stirling.software.proprietary.pdf.FlexibleCSVWriter;
import technology.tabula.ObjectExtractor;
import technology.tabula.Page;
import technology.tabula.Table;
import technology.tabula.extractors.SpreadsheetExtractionAlgorithm;
@Slf4j
@Service
@RequiredArgsConstructor
public class PdfContentExtractor {
private final TabulaTableParser tabulaTableParser;
private final PdfIngester pdfIngester;
private static final int MAX_CHARACTERS_PER_PAGE = 4_000;
private static final int TEXT_PRESENCE_THRESHOLD = 20;
@@ -101,21 +106,21 @@ public class PdfContentExtractor {
* @return list of CSV strings (one per table), empty if no tables found
*/
public List<String> extractTablesAsCsv(PDDocument document, int pageNumber) throws IOException {
SpreadsheetExtractionAlgorithm sea = new SpreadsheetExtractionAlgorithm();
List<TableFragment> fragments = tabulaTableParser.parse(document, pageNumber);
if (fragments.isEmpty()) return List.of();
CSVFormat format =
CSVFormat.EXCEL.builder().setEscape('"').setQuoteMode(QuoteMode.ALL).build();
List<String> csvStrings = new ArrayList<>();
try (ObjectExtractor extractor = new ObjectExtractor(document)) {
Page tabulaPage = extractor.extract(pageNumber);
List<Table> tables = sea.extract(tabulaPage);
for (Table table : tables) {
StringWriter sw = new StringWriter();
FlexibleCSVWriter csvWriter = new FlexibleCSVWriter(format);
csvWriter.write(sw, Collections.singletonList(table));
csvStrings.add(sw.toString());
for (TableFragment fragment : fragments) {
StringWriter sw = new StringWriter();
try (CSVPrinter printer = format.print(sw)) {
for (List<String> row : fragment.rawRows()) {
printer.printRecord(row);
}
}
csvStrings.add(sw.toString());
}
return csvStrings;
}
@@ -183,6 +188,8 @@ public class PdfContentExtractor {
case PAGE_TEXT, FULL_TEXT ->
Optional.<PdfContentResult>ofNullable(
extractText(lf, fileReq, remainingPages, remainingCharacters));
case PAGE_LAYOUT ->
Optional.<PdfContentResult>ofNullable(extractPageLayout(lf, remainingPages));
default -> {
log.warn(
"Content type {} not yet implemented, skipping for {}",
@@ -207,6 +214,35 @@ public class PdfContentExtractor {
return extracted.isEmpty() ? null : buildExtractedFileText(lf.fileName(), extracted);
}
private PageLayoutFileResult extractPageLayout(LoadedFile lf, int maxPages) throws IOException {
List<ParsedPage> parsedPages = pdfIngester.parse(lf.document(), maxPages);
List<LayoutPage> pages = new ArrayList<>();
for (ParsedPage pp : parsedPages) {
if (pp.layoutLines().isEmpty()) continue;
List<LayoutLine> lines = new ArrayList<>();
for (RawLine rawLine : pp.layoutLines()) {
List<LayoutFragment> fragments = new ArrayList<>();
for (TextFragment tf : rawLine.fragments()) {
fragments.add(
new LayoutFragment(
tf.text(),
tf.bounds().x(),
tf.bounds().y(),
tf.bounds().width(),
tf.fontSize(),
tf.bold()));
}
lines.add(new LayoutLine(rawLine.bounds().y(), fragments));
}
pages.add(new LayoutPage(pp.pageNumber(), lines));
}
if (pages.isEmpty()) return null;
PageLayoutFileResult result = new PageLayoutFileResult();
result.setFileName(lf.fileName());
result.setPages(pages);
return result;
}
private WorkflowArtifact buildArtifact(ArtifactKind kind, List<PdfContentResult> results) {
return switch (kind) {
case EXTRACTED_TEXT -> {
@@ -214,10 +250,12 @@ public class PdfContentExtractor {
artifact.setFiles(results.stream().map(ExtractedFileText.class::cast).toList());
yield artifact;
}
case PAGE_LAYOUT -> {
PageLayoutArtifact artifact = new PageLayoutArtifact();
artifact.setFiles(results.stream().map(PageLayoutFileResult.class::cast).toList());
yield artifact;
}
case TOOL_REPORT ->
// TOOL_REPORT artifacts don't come from PDF content extraction — they're
// built by AiWorkflowService from tool-response metadata. Never reached
// from this code path; presence in the enum is to satisfy the switch.
throw new IllegalArgumentException(
"TOOL_REPORT artifacts are not produced by PdfContentExtractor");
};
@@ -331,6 +369,7 @@ public class PdfContentExtractor {
*/
enum ArtifactKind {
EXTRACTED_TEXT("extracted_text"),
PAGE_LAYOUT("page_layout"),
TOOL_REPORT("tool_report");
private final String value;
@@ -394,4 +433,40 @@ public class PdfContentExtractor {
this.report = report;
}
}
// Serialization contract with the Python engine — see PageLayoutArtifactContractTest.
/** One text fragment with its bounding-box geometry and font properties. */
record LayoutFragment(
String text, float x, float y, float width, float fontSize, boolean bold) {}
/** A visual line on the page: y-coordinate and all fragments on that line. */
record LayoutLine(float y, List<LayoutFragment> fragments) {}
/** All layout lines for a single page. */
record LayoutPage(int pageNumber, List<LayoutLine> lines) {}
/** Page layout data for one file, as a PdfContentResult. */
@Data
static final class PageLayoutFileResult implements PdfContentResult {
private String fileName;
private List<LayoutPage> pages = new ArrayList<>();
@Override
public ArtifactKind getArtifactKind() {
return ArtifactKind.PAGE_LAYOUT;
}
@Override
public int pagesConsumed() {
return pages.size();
}
}
/** Artifact carrying full spatial page layout for all input files. */
@Data
static final class PageLayoutArtifact implements WorkflowArtifact {
private final ArtifactKind kind = ArtifactKind.PAGE_LAYOUT;
private List<PageLayoutFileResult> files = new ArrayList<>();
}
}