Development: Reuse compliance analysis across languages via snippet mapping

openapi/openapi.yaml

@@ -215,6 +215,23 @@ paths:
               schema:
                 type: array
                 items: {type: string}
+  /api/ai/map-compliance-issues:
+    post:
+      tags: [ai-resource]
+      operationId: mapComplianceIssues
+      requestBody:
+        content:
+          application/json:
+            schema: {$ref: '#/components/schemas/MapComplianceIssuesRequestDTO'}
+        required: true
+      responses:
+        '200':
+          description: OK
+          content:
+            application/json:
+              schema:
+                type: array
+                items: {$ref: '#/components/schemas/ComplianceIssue'}
   /api/ai/translateJobDescriptionStream:
     put:
       tags: [ai-resource]
@@ -3880,6 +3897,16 @@ components:
         email: {type: string, format: email, minLength: 1}
         password: {type: string, minLength: 1}
       required: [email, password]
+    MapComplianceIssuesRequestDTO:
+      type: object
+      properties:
+        complianceIssues:
+          type: array
+          items: {$ref: '#/components/schemas/ComplianceIssue'}
+        jobId: {type: string, format: uuid}
+        toLang: {type: string}
+        translatedText: {type: string, minLength: 1}
+      required: [complianceIssues, translatedText]
     MultipartUploadRequest:
       type: object
       properties:

src/main/java/de/tum/cit/aet/ai/dto/MapComplianceIssuesRequestDTO.java

@@ -0,0 +1,16 @@
+package de.tum.cit.aet.ai.dto;
+
+import com.fasterxml.jackson.annotation.JsonInclude;
+import de.tum.cit.aet.ai.domain.ComplianceIssue;
+import jakarta.validation.constraints.NotBlank;
+import jakarta.validation.constraints.NotNull;
+import java.util.List;
+import java.util.UUID;
+
+@JsonInclude
+public record MapComplianceIssuesRequestDTO(
+    String toLang,
+    UUID jobId,
+    @NotBlank String translatedText,
+    @NotNull List<ComplianceIssue> complianceIssues
+) {}

src/main/java/de/tum/cit/aet/ai/service/AiService.java

@@ -6,6 +6,8 @@
 import de.tum.cit.aet.ai.domain.ComplianceIssue;
 import de.tum.cit.aet.ai.dto.ExtractedApplicationDataDTO;
 import de.tum.cit.aet.ai.dto.ExtractedCertificateDataDTO;
+import de.tum.cit.aet.ai.dto.MapComplianceIssuesRequestDTO;
+import de.tum.cit.aet.ai.util.SnippetMatcher;
 import de.tum.cit.aet.application.service.ApplicationService;
 import de.tum.cit.aet.core.documents.service.DocumentService;
 import de.tum.cit.aet.core.dto.GenderBiasAnalysisResponse;
@@ -27,6 +29,7 @@
 import java.util.Set;
 import java.util.UUID;
 import java.util.concurrent.atomic.AtomicReference;
+import java.util.stream.Collectors;
 import javax.imageio.ImageIO;
 import lombok.extern.slf4j.Slf4j;
 import org.apache.pdfbox.Loader;
@@ -74,6 +77,9 @@ public class AiService {
     @Value("classpath:prompts/AnalyzeComplianceText.st")
     private Resource complianceResource;
 
+    @Value("classpath:prompts/SnippetMapping.st")
+    private Resource snippetMappingResource;
+
     private final ChatClient chatClient;
 
     private final JobService jobService;
@@ -496,4 +502,72 @@ public List<ComplianceIssue> analyzeJobDescription(
 
         return complianceIssues;
     }
+
+    /**
+     * Maps the snippets of an existing source-language compliance analysis onto the
+     * translated job description, avoiding a second full LLM compliance analysis.
+     *
+     * @param request DTO containing the source compliance issues, translated text, target language, and job ID
+     * @return the persisted list of mapped issues, in the same order as sourceIssues
+     */
+    public List<ComplianceIssue> mapComplianceIssues(MapComplianceIssuesRequestDTO request) {
+        // Empty source issues mean "no issues found" -> clear stale target-language issues.
+        if (request.complianceIssues().isEmpty()) {
+            jobService.updateComplianceIssues(request.jobId(), List.of(), request.toLang());
+            return List.of();
+        }
+
+        String snippets = java.util.stream.IntStream.range(0, request.complianceIssues().size())
+            .mapToObj(index -> (index + 1) + "\t" + request.complianceIssues().get(index).getText().trim())
+            .collect(Collectors.joining("\n"));
+
+        List<String> mappedTexts;
+        try {
+            mappedTexts = chatClient
+                .prompt()
+                .user(u ->
+                    u
+                        .text(snippetMappingResource)
+                        .param("count", String.valueOf(request.complianceIssues().size()))
+                        .param("snippets", snippets)
+                        .param("translatedText", request.translatedText())
+                )
+                .call()
+                .entity(new ParameterizedTypeReference<List<String>>() {});
+            aiFeatureToggleService.recordSuccess();
+        } catch (Exception e) {
+            aiFeatureToggleService.recordFailure();
+            throw new InternalServerException("Compliance issue mapping failed", e);
+        }
+
+        if (mappedTexts == null || mappedTexts.size() != request.complianceIssues().size()) {
+            aiFeatureToggleService.recordFailure();
+            throw new InternalServerException("Mapping returned an invalid number of snippets");
+        }
+
+        List<ComplianceIssue> mappedIssues = new ArrayList<>();
+        for (int i = 0; i < request.complianceIssues().size(); i++) {
+            String mappedText = mappedTexts.get(i);
+            String mapped = mappedText == null ? null : mappedText.trim();
+            if (!SnippetMatcher.isVerbatim(request.translatedText(), mapped)) {
+                log.warn("Snippet {} not found in translated text, dropping", i);
+                continue;
+            }
+            ComplianceIssue sourceIssue = request.complianceIssues().get(i);
+            mappedIssues.add(
+                new ComplianceIssue(
+                    sourceIssue.getId(),
+                    sourceIssue.getCategory(),
+                    mapped,
+                    sourceIssue.getArticle(),
+                    sourceIssue.getExplanation(),
+                    sourceIssue.getAction(),
+                    request.toLang()
+                )
+            );
+        }
+
+        jobService.updateComplianceIssues(request.jobId(), mappedIssues, request.toLang());
+        return mappedIssues;
+    }
 }

src/main/java/de/tum/cit/aet/ai/util/SnippetMatcher.java

@@ -0,0 +1,22 @@
+package de.tum.cit.aet.ai.util;
+
+/**
+ * Validates mapped compliance snippets against the translated target text.
+ */
+public final class SnippetMatcher {
+
+    private SnippetMatcher() {}
+
+    /**
+     * Checks whether a non-empty candidate occurs verbatim in the target text.
+     * Matching is case-sensitive because the model copies the phrase verbatim and
+     * the client searches for that exact phrase in the editor.
+     *
+     * @param targetText translated job description
+     * @param candidate mapped compliance snippet
+     * @return {@code true} when the candidate is non-empty and occurs verbatim
+     */
+    public static boolean isVerbatim(String targetText, String candidate) {
+        return candidate != null && !candidate.isEmpty() && targetText.contains(candidate);
+    }
+}

src/main/java/de/tum/cit/aet/ai/web/AiResource.java

@@ -2,13 +2,16 @@
 
 import de.tum.cit.aet.ai.domain.ComplianceIssue;
 import de.tum.cit.aet.ai.dto.ExtractedApplicationDataDTO;
+import de.tum.cit.aet.ai.dto.MapComplianceIssuesRequestDTO;
 import de.tum.cit.aet.ai.dto.TranslateComplianceDTO;
 import de.tum.cit.aet.ai.service.AiFeatureToggleService;
 import de.tum.cit.aet.ai.service.AiService;
 import de.tum.cit.aet.core.security.annotations.ApplicantOrAdmin;
 import de.tum.cit.aet.core.security.annotations.ProfessorOrEmployeeOrAdmin;
 import de.tum.cit.aet.job.dto.JobFormDTO;
+import jakarta.validation.Valid;
 import java.util.List;
+import java.util.UUID;
 import lombok.extern.slf4j.Slf4j;
 import org.springframework.context.annotation.Profile;
 import org.springframework.http.HttpStatus;
@@ -70,7 +73,7 @@ public ResponseEntity<Flux<String>> generateJobApplicationDraftStream(
     @PutMapping(value = "translateJobDescriptionStream", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
     public ResponseEntity<Flux<String>> translateJobDescriptionStream(
         @RequestParam("toLang") String toLang,
-        @RequestBody TranslateComplianceDTO request
+        @Valid @RequestBody TranslateComplianceDTO request
     ) {
         if (!aiFeatureToggleService.isAiAvailable()) {
             return ResponseEntity.status(HttpStatus.SERVICE_UNAVAILABLE).build();
@@ -79,6 +82,22 @@ public ResponseEntity<Flux<String>> translateJobDescriptionStream(
         return ResponseEntity.ok(aiService.translateTextStream(request.text(), toLang));
     }
 
+    /**
+     * Maps compliance text snippets from original lang to target lang during stream-translate.
+     *
+     * @param request A DTO containing the text to translate
+     * @return a ResponseEntity of mapped snippets for target compliance analysis
+     */
+    @ProfessorOrEmployeeOrAdmin
+    @PostMapping(value = "map-compliance-issues", produces = MediaType.APPLICATION_JSON_VALUE)
+    public ResponseEntity<List<ComplianceIssue>> mapComplianceIssues(@Valid @RequestBody MapComplianceIssuesRequestDTO request) {
+        if (!aiFeatureToggleService.isAiAvailable()) {
+            return ResponseEntity.status(HttpStatus.SERVICE_UNAVAILABLE).build();
+        }
+        log.info("POST /api/ai/map-compliance-issues - Compliance snippet-mapping request received (toLang={})", request.toLang());
+        return ResponseEntity.ok(aiService.mapComplianceIssues(request));
+    }
+
     /**
      * Extracts applicant data from PDF files using AI and persists the extracted
      * values into the application entity.
@@ -132,7 +151,7 @@ public ResponseEntity<List<ComplianceIssue>> analyzeJobDescriptionForCompliance(
         @RequestParam(defaultValue = "en") String userLanguage
     ) {
         // Service skips LLM calls internally when AI is disabled, rule-based gender bias analysis and score computation remain enabled
-        log.info("POST /api/ai/analyzeJobDescription - Request received (toLang={})", descriptionLanguage);
+        log.info("POST /api/ai/analyzeJobDescription - Compliance analysis request received (toLang={})", descriptionLanguage);
         return ResponseEntity.ok(aiService.analyzeCurrentJobDescription(jobForm, descriptionLanguage, userLanguage));
     }
 }

src/main/java/de/tum/cit/aet/job/service/JobService.java

@@ -39,6 +39,7 @@
 import java.util.Objects;
 import java.util.Set;
 import java.util.UUID;
+import java.util.function.Consumer;
 import java.util.stream.Collectors;
 import lombok.RequiredArgsConstructor;
 import org.springframework.data.domain.Page;
@@ -543,30 +544,59 @@ public void updateJobDescriptionLanguage(String jobId, String toLang, String tra
     }
 
     /**
-     * Updates AI-generated analysis fields for a job.
+     * Updates AI-generated analysis fields for a job: replaces the compliance
+     * issues for the given language and overwrites the combined gender bias score.
      *
-     * @param jobId the job identifier
-     * @param score the combined AI score to persist
-     * @param complianceAnalysis the compliance issues detected for the job description
-     * @param lang the language for which existing issues should be replaced
+     * @param jobId               the job identifier
+     * @param score               the combined AI score to persist
+     * @param complianceAnalysis  compliance issues detected for the given language
+     * @param lang                the analyzed language ("de" or "en")
      */
     public void updateAiAnalysis(UUID jobId, int score, List<ComplianceIssue> complianceAnalysis, String lang) {
+        applyJobChangeForAnalysis(jobId, job -> {
+            replaceComplianceIssuesForLanguage(job, complianceAnalysis, lang);
+            job.setGenderBiasScore(score);
+        });
+    }
+
+    /**
+     * Replaces the compliance issues for a single language without touching the
+     * gender bias score. Used by the snippet-mapping flow, where the score has
+     * already been written by the source-language analysis and must not be reset.
+     *
+     * @param jobId               the job identifier
+     * @param complianceAnalysis  compliance issues for the target language
+     * @param lang                the target language ("de" or "en")
+     */
+    public void updateComplianceIssues(UUID jobId, List<ComplianceIssue> complianceAnalysis, String lang) {
+        applyJobChangeForAnalysis(jobId, job -> replaceComplianceIssuesForLanguage(job, complianceAnalysis, lang));
+    }
+
+    /**
+     * Loads the job, applies the given change, and persists in a single repository write.
+     */
+    private void applyJobChangeForAnalysis(UUID jobId, Consumer<Job> changes) {
         if (jobId == null) {
             return;
         }
-
         Job job = jobRepository.findByIdWithCompliance(jobId).orElseThrow(() -> EntityNotFoundException.forId("Job", jobId));
+        currentUserService.isAdminOrMemberOf(job.getResearchGroup());
+        changes.accept(job);
+        jobRepository.save(job);
+    }
 
-        // Keep issues from the other language, add new ones for target language
-        List issuesToSave = job
+    /**
+     * Replaces compliance issues for the given language.
+     * Issues from other languages stay unchanged.
+     * Updates the job in place and caller saves it.
+     */
+    private void replaceComplianceIssuesForLanguage(Job job, List<ComplianceIssue> complianceAnalysis, String lang) {
+        List<ComplianceIssue> issuesToSave = job
             .getComplianceIssues()
             .stream()
             .filter(issue -> !Objects.equals(issue.getLanguage(), lang))
             .collect(Collectors.toCollection(ArrayList::new));
-
         issuesToSave.addAll(complianceAnalysis);
-        job.setGenderBiasScore(score);
         job.setComplianceIssues(issuesToSave);
-        jobRepository.save(job);
     }
 }