Wired optional LLM extraction into the analysis path

src/repo_registry/core/service.py

@@ -17,6 +17,8 @@ from repo_registry.core.models import (
 )
 from repo_registry.candidate_graph.generator import CandidateGraphGenerator
 from repo_registry.content_indexing.extractor import ContentExtractor
+from repo_registry.llm_extraction.extractor import LLMCandidateExtractor
+from repo_registry.llm_extraction.mapper import LLMExtractionMapper
 from repo_registry.repo_ingestion.git import GitIngestionService
 from repo_registry.repo_ingestion.metadata import RepositoryMetadataExtractor
 from repo_registry.repo_scanning.scanner import DeterministicScanner
@@ -30,6 +32,7 @@ class RegistryService:
         self,
         store: RegistryStore,
         ingestion: GitIngestionService | None = None,
+        llm_extractor: LLMCandidateExtractor | None = None,
     ) -> None:
         self.store = store
         self.scanner = DeterministicScanner()
@@ -37,6 +40,8 @@ class RegistryService:
         self.metadata_extractor = RepositoryMetadataExtractor()
         self.candidate_generator = CandidateGraphGenerator()
         self.content_extractor = ContentExtractor()
+        self.llm_extractor = llm_extractor
+        self.llm_mapper = LLMExtractionMapper()
 
     def register_repository(
         self,
@@ -122,11 +127,7 @@ class RegistryService:
             chunks,
         )
         stored_chunks = self.store.list_content_chunks(repository_id, completed_run.id)
-        candidates = self.candidate_generator.generate(
-            repository,
-            facts,
-            stored_chunks,
-        )
+        candidates = self._generate_candidates(repository, facts, stored_chunks)
         self.store.replace_candidate_graph(repository_id, completed_run.id, candidates)
         return ScanSummary(
             analysis_run=completed_run,
@@ -134,6 +135,18 @@ class RegistryService:
             facts=facts,
         )
 
+    def _generate_candidates(
+        self,
+        repository: Repository,
+        facts: list[ObservedFact],
+        chunks: list[ContentChunk],
+    ):
+        if self.llm_extractor is not None:
+            extracted = self.llm_extractor.extract(repository, chunks)
+            if extracted:
+                return self.llm_mapper.map(extracted, facts, chunks)
+        return self.candidate_generator.generate(repository, facts, chunks)
+
     def list_analysis_runs(self, repository_id: int) -> list[AnalysisRun]:
         return self.store.list_analysis_runs(repository_id)