docs: update CLAUDE.md references from Vertex AI to google-genai (refs #231)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

ocr/app/CLAUDE.md

@@ -7,7 +7,7 @@ Python OCR microservice (FastAPI). Primary engine: PaddleOCR PP-OCRv4 with optio
 | File | What | When to read |
 | ---- | ---- | ------------ |
 | `main.py` | FastAPI application entry point | Route registration, app setup |
-| `config.py` | Configuration settings (OCR engines, Vertex AI, Redis, Vision API limits) | Environment variables, settings |
+| `config.py` | Configuration settings (OCR engines, Google GenAI, Redis, Vision API limits) | Environment variables, settings |
 | `__init__.py` | Package init | Package structure |
 
 ## Subdirectories

ocr/app/config.py

@@ -29,7 +29,7 @@ class Settings:
             os.getenv("VISION_MONTHLY_LIMIT", "1000")
         )
 
-        # Vertex AI / Gemini configuration
+        # Google GenAI / Gemini configuration
         self.vertex_ai_project: str = os.getenv("VERTEX_AI_PROJECT", "")
         self.vertex_ai_location: str = os.getenv(
             "VERTEX_AI_LOCATION", "global"

ocr/app/engines/CLAUDE.md

@@ -3,7 +3,7 @@
 OCR engine abstraction layer. Two categories of engines:
 
 1. **OcrEngine subclasses** (image-to-text): PaddleOCR, Google Vision, Hybrid. Accept image bytes, return text + confidence + word boxes.
-2. **GeminiEngine** (PDF-to-structured-data and VIN decode): Standalone module for maintenance schedule extraction and VIN decoding via Vertex AI. Accepts PDF bytes or VIN strings, returns structured JSON. Not an OcrEngine subclass because the interface signatures differ.
+2. **GeminiEngine** (PDF-to-structured-data and VIN decode): Standalone module for maintenance schedule extraction and VIN decoding via google-genai SDK. Accepts PDF bytes or VIN strings, returns structured JSON. Not an OcrEngine subclass because the interface signatures differ.
 
 ## Files
 
@@ -15,7 +15,7 @@ OCR engine abstraction layer. Two categories of engines:
 | `cloud_engine.py` | Google Vision TEXT_DETECTION fallback engine (WIF authentication) | Cloud OCR configuration, API quota |
 | `hybrid_engine.py` | Combines primary + fallback engine with confidence threshold switching | Engine selection logic, fallback behavior |
 | `engine_factory.py` | Factory function and engine registry for instantiation | Adding new engine types |
-| `gemini_engine.py` | Gemini 2.5 Flash integration for maintenance schedule extraction and VIN decoding (Vertex AI SDK, 20MB PDF limit, structured JSON output) | Manual extraction debugging, VIN decode, Gemini configuration |
+| `gemini_engine.py` | Gemini 2.5 Flash integration for maintenance schedule extraction and VIN decoding (google-genai SDK, 20MB PDF limit, structured JSON output, Google Search grounding for VIN decode) | Manual extraction debugging, VIN decode, Gemini configuration |
 
 ## Engine Selection
 

ocr/app/engines/gemini_engine.py

@@ -2,7 +2,7 @@
 
 Standalone module (does NOT extend OcrEngine) because Gemini performs
 semantic document understanding, not traditional OCR word-box extraction.
-Uses Vertex AI SDK with structured JSON output enforcement.
+Uses google-genai SDK with structured JSON output enforcement.
 """
 
 import json
@@ -117,34 +117,34 @@ Return the vehicle's make, model, trim level, body type, drive type, fuel type,
 """
 
 _VIN_DECODE_SCHEMA: dict[str, Any] = {
-    "type": "object",
+    "type": "OBJECT",
     "properties": {
-        "year": {"type": "integer", "nullable": True},
+        "year": {"type": "INTEGER", "nullable": True},
-        "make": {"type": "string", "nullable": True},
+        "make": {"type": "STRING", "nullable": True},
-        "model": {"type": "string", "nullable": True},
+        "model": {"type": "STRING", "nullable": True},
-        "trimLevel": {"type": "string", "nullable": True},
+        "trimLevel": {"type": "STRING", "nullable": True},
-        "bodyType": {"type": "string", "nullable": True},
+        "bodyType": {"type": "STRING", "nullable": True},
-        "driveType": {"type": "string", "nullable": True},
+        "driveType": {"type": "STRING", "nullable": True},
-        "fuelType": {"type": "string", "nullable": True},
+        "fuelType": {"type": "STRING", "nullable": True},
-        "engine": {"type": "string", "nullable": True},
+        "engine": {"type": "STRING", "nullable": True},
-        "transmission": {"type": "string", "nullable": True},
+        "transmission": {"type": "STRING", "nullable": True},
-        "confidence": {"type": "number"},
+        "confidence": {"type": "NUMBER"},
     },
     "required": ["confidence"],
 }
 
 _RESPONSE_SCHEMA: dict[str, Any] = {
-    "type": "object",
+    "type": "OBJECT",
     "properties": {
         "maintenanceSchedule": {
-            "type": "array",
+            "type": "ARRAY",
             "items": {
-                "type": "object",
+                "type": "OBJECT",
                 "properties": {
-                    "serviceName": {"type": "string"},
+                    "serviceName": {"type": "STRING"},
-                    "intervalMiles": {"type": "number", "nullable": True},
+                    "intervalMiles": {"type": "NUMBER", "nullable": True},
-                    "intervalMonths": {"type": "number", "nullable": True},
+                    "intervalMonths": {"type": "NUMBER", "nullable": True},
-                    "details": {"type": "string", "nullable": True},
+                    "details": {"type": "STRING", "nullable": True},
                 },
                 "required": ["serviceName"],
             },
@@ -206,20 +206,21 @@ class GeminiEngine:
     Standalone class (not an OcrEngine subclass) because Gemini performs
     semantic document understanding rather than traditional OCR.
 
-    Uses lazy initialization: the Vertex AI client is not created until
+    Uses lazy initialization: the Gemini client is not created until
     the first call to ``extract_maintenance()`` or ``decode_vin()``.
     """
 
     def __init__(self) -> None:
-        self._model: Any | None = None
+        self._client: Any | None = None
+        self._model_name: str = ""
 
-    def _get_model(self) -> Any:
+    def _get_client(self) -> Any:
-        """Create the GenerativeModel on first use.
+        """Create the genai.Client on first use.
 
         Authentication uses the same WIF credential path as Google Vision.
         """
-        if self._model is not None:
+        if self._client is not None:
-            return self._model
+            return self._client
 
         key_path = settings.google_vision_key_path
         if not os.path.isfile(key_path):
@@ -229,46 +230,37 @@ class GeminiEngine:
             )
 
         try:
-            from google.cloud import aiplatform  # type: ignore[import-untyped]
+            from google import genai  # type: ignore[import-untyped]
-            from vertexai.generative_models import (  # type: ignore[import-untyped]
-                GenerationConfig,
-                GenerativeModel,
-            )
 
-            # Point ADC at the WIF credential config
+            # Point ADC at the WIF credential config (must be set BEFORE Client construction)
             os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = key_path
             os.environ["GOOGLE_EXTERNAL_ACCOUNT_ALLOW_EXECUTABLES"] = "1"
 
-            aiplatform.init(
+            self._client = genai.Client(
+                vertexai=True,
                 project=settings.vertex_ai_project,
                 location=settings.vertex_ai_location,
             )
-
+            self._model_name = settings.gemini_model
-            model_name = settings.gemini_model
-            self._model = GenerativeModel(model_name)
-            self._generation_config = GenerationConfig(
-                response_mime_type="application/json",
-                response_schema=_RESPONSE_SCHEMA,
-            )
 
             logger.info(
                 "Gemini engine initialized (model=%s, project=%s, location=%s)",
-                model_name,
+                self._model_name,
                 settings.vertex_ai_project,
                 settings.vertex_ai_location,
             )
-            return self._model
+            return self._client
 
         except ImportError as exc:
-            logger.exception("Vertex AI SDK import failed")
+            logger.exception("google-genai SDK import failed")
             raise GeminiUnavailableError(
-                "google-cloud-aiplatform is not installed. "
+                "google-genai is not installed. "
-                "Install with: pip install google-cloud-aiplatform"
+                "Install with: pip install google-genai"
             ) from exc
         except Exception as exc:
-            logger.exception("Vertex AI authentication failed")
+            logger.exception("Gemini authentication failed: %s", type(exc).__name__)
             raise GeminiUnavailableError(
-                f"Vertex AI authentication failed: {exc}"
+                f"Gemini authentication failed: {exc}"
             ) from exc
 
     def extract_maintenance(
@@ -293,19 +285,23 @@ class GeminiEngine:
                 "inline processing. Upload to GCS and use a gs:// URI instead."
             )
 
-        model = self._get_model()
+        client = self._get_client()
 
         try:
-            from vertexai.generative_models import Part  # type: ignore[import-untyped]
+            from google.genai import types  # type: ignore[import-untyped]
 
-            pdf_part = Part.from_data(
+            pdf_part = types.Part.from_bytes(
                 data=pdf_bytes,
                 mime_type="application/pdf",
             )
 
-            response = model.generate_content(
+            response = client.models.generate_content(
-                [pdf_part, _EXTRACTION_PROMPT],
+                model=self._model_name,
-                generation_config=self._generation_config,
+                contents=[pdf_part, _EXTRACTION_PROMPT],
+                config=types.GenerateContentConfig(
+                    response_mime_type="application/json",
+                    response_schema=_RESPONSE_SCHEMA,
+                ),
             )
 
             raw = json.loads(response.text)
@@ -358,7 +354,7 @@ class GeminiEngine:
             GeminiProcessingError: If Gemini fails to decode the VIN.
             GeminiUnavailableError: If the engine cannot be initialized.
         """
-        model = self._get_model()
+        client = self._get_client()
 
         # Resolve year deterministically from VIN structure
         resolved_year = resolve_vin_year(vin)
@@ -371,21 +367,21 @@ class GeminiEngine:
         )
 
         try:
-            from vertexai.generative_models import GenerationConfig  # type: ignore[import-untyped]
+            from google.genai import types  # type: ignore[import-untyped]
-
-            vin_config = GenerationConfig(
-                response_mime_type="application/json",
-                response_schema=_VIN_DECODE_SCHEMA,
-            )
 
             prompt = _VIN_DECODE_PROMPT.format(
                 vin=vin,
                 year=resolved_year or "unknown",
                 year_code=year_code,
             )
-            response = model.generate_content(
+            response = client.models.generate_content(
-                [prompt],
+                model=self._model_name,
-                generation_config=vin_config,
+                contents=[prompt],
+                config=types.GenerateContentConfig(
+                    response_mime_type="application/json",
+                    response_schema=_VIN_DECODE_SCHEMA,
+                    tools=[types.Tool(google_search=types.GoogleSearch())],
+                ),
             )
 
             raw = json.loads(response.text)

ocr/app/extractors/maintenance_receipt_extractor.py

@@ -14,6 +14,7 @@ import time
 from typing import Any, Optional
 
 from app.config import settings
+from app.engines.gemini_engine import GeminiUnavailableError
 from app.extractors.receipt_extractor import (
     ExtractedField,
     ReceiptExtractionResult,
@@ -54,16 +55,16 @@ OCR Text:
 """
 
 _RECEIPT_RESPONSE_SCHEMA: dict[str, Any] = {
-    "type": "object",
+    "type": "OBJECT",
     "properties": {
-        "serviceName": {"type": "string", "nullable": True},
+        "serviceName": {"type": "STRING", "nullable": True},
-        "serviceDate": {"type": "string", "nullable": True},
+        "serviceDate": {"type": "STRING", "nullable": True},
-        "totalCost": {"type": "number", "nullable": True},
+        "totalCost": {"type": "NUMBER", "nullable": True},
-        "shopName": {"type": "string", "nullable": True},
+        "shopName": {"type": "STRING", "nullable": True},
-        "laborCost": {"type": "number", "nullable": True},
+        "laborCost": {"type": "NUMBER", "nullable": True},
-        "partsCost": {"type": "number", "nullable": True},
+        "partsCost": {"type": "NUMBER", "nullable": True},
-        "odometerReading": {"type": "number", "nullable": True},
+        "odometerReading": {"type": "NUMBER", "nullable": True},
-        "vehicleInfo": {"type": "string", "nullable": True},
+        "vehicleInfo": {"type": "STRING", "nullable": True},
     },
     "required": [
         "serviceName",
@@ -87,8 +88,8 @@ class MaintenanceReceiptExtractor:
     """
 
     def __init__(self) -> None:
-        self._model: Any | None = None
+        self._client: Any | None = None
-        self._generation_config: Any | None = None
+        self._model_name: str = ""
 
     def extract(
         self,
@@ -169,47 +170,52 @@ class MaintenanceReceiptExtractor:
             processing_time_ms=processing_time_ms,
         )
 
-    def _get_model(self) -> Any:
+    def _get_client(self) -> Any:
-        """Lazy-initialize Vertex AI Gemini model.
+        """Lazy-initialize google-genai Gemini client.
 
         Uses the same authentication pattern as GeminiEngine.
         """
-        if self._model is not None:
+        if self._client is not None:
-            return self._model
+            return self._client
 
         key_path = settings.google_vision_key_path
         if not os.path.isfile(key_path):
-            raise RuntimeError(
+            raise GeminiUnavailableError(
                 f"Google credential config not found at {key_path}. "
                 "Set GOOGLE_VISION_KEY_PATH or mount the secret."
             )
 
-        from google.cloud import aiplatform  # type: ignore[import-untyped]
+        try:
-        from vertexai.generative_models import (  # type: ignore[import-untyped]
+            from google import genai  # type: ignore[import-untyped]
-            GenerationConfig,
-            GenerativeModel,
-        )
 
+            # Point ADC at the WIF credential config (must be set BEFORE Client construction)
             os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = key_path
             os.environ["GOOGLE_EXTERNAL_ACCOUNT_ALLOW_EXECUTABLES"] = "1"
 
-        aiplatform.init(
+            self._client = genai.Client(
+                vertexai=True,
                 project=settings.vertex_ai_project,
                 location=settings.vertex_ai_location,
             )
-
+            self._model_name = settings.gemini_model
-        model_name = settings.gemini_model
-        self._model = GenerativeModel(model_name)
-        self._generation_config = GenerationConfig(
-            response_mime_type="application/json",
-            response_schema=_RECEIPT_RESPONSE_SCHEMA,
-        )
 
             logger.info(
-            "Maintenance receipt Gemini model initialized (model=%s)",
+                "Maintenance receipt Gemini client initialized (model=%s)",
-            model_name,
+                self._model_name,
             )
-        return self._model
+            return self._client
+
+        except ImportError as exc:
+            logger.exception("google-genai SDK import failed")
+            raise GeminiUnavailableError(
+                "google-genai is not installed. "
+                "Install with: pip install google-genai"
+            ) from exc
+        except Exception as exc:
+            logger.exception("Gemini authentication failed: %s", type(exc).__name__)
+            raise GeminiUnavailableError(
+                f"Gemini authentication failed: {exc}"
+            ) from exc
 
     def _extract_with_gemini(self, ocr_text: str) -> dict:
         """Send OCR text to Gemini for semantic field extraction.
@@ -220,13 +226,19 @@ class MaintenanceReceiptExtractor:
         Returns:
             Dictionary of field_name -> extracted_value from Gemini.
         """
-        model = self._get_model()
+        client = self._get_client()
+
+        from google.genai import types  # type: ignore[import-untyped]
 
         prompt = _RECEIPT_EXTRACTION_PROMPT.format(ocr_text=ocr_text)
 
-        response = model.generate_content(
+        response = client.models.generate_content(
-            [prompt],
+            model=self._model_name,
-            generation_config=self._generation_config,
+            contents=[prompt],
+            config=types.GenerateContentConfig(
+                response_mime_type="application/json",
+                response_schema=_RECEIPT_RESPONSE_SCHEMA,
+            ),
         )
 
         raw = json.loads(response.text)

ocr/tests/test_gemini_engine.py

@@ -2,11 +2,11 @@
 
 Covers: GeminiEngine initialization, PDF size validation,
 successful extraction, empty results, and error handling.
-All Vertex AI SDK calls are mocked.
+All google-genai SDK calls are mocked.
 """
 
 import json
-from unittest.mock import MagicMock, patch, PropertyMock
+from unittest.mock import MagicMock, patch
 
 import pytest
 
@@ -156,22 +156,16 @@ class TestExtractMaintenance:
             },
         ]
 
-        mock_model = MagicMock()
+        mock_client = MagicMock()
-        mock_model.generate_content.return_value = _make_gemini_response(schedule)
+        mock_client.models.generate_content.return_value = _make_gemini_response(schedule)
 
-        with (
+        with patch.dict("sys.modules", {
-            patch(
+            "google.genai": MagicMock(),
-                "app.engines.gemini_engine.importlib_vertex_ai"
+            "google.genai.types": MagicMock(),
-            ) if False else patch.dict("sys.modules", {
+        }):
-                "google.cloud": MagicMock(),
-                "google.cloud.aiplatform": MagicMock(),
-                "vertexai": MagicMock(),
-                "vertexai.generative_models": MagicMock(),
-            }),
-        ):
             engine = GeminiEngine()
-            engine._model = mock_model
+            engine._client = mock_client
-            engine._generation_config = MagicMock()
+            engine._model_name = "gemini-2.5-flash"
 
             result = engine.extract_maintenance(_make_pdf_bytes())
 
@@ -200,12 +194,12 @@ class TestExtractMaintenance:
         mock_settings.vertex_ai_location = "us-central1"
         mock_settings.gemini_model = "gemini-2.5-flash"
 
-        mock_model = MagicMock()
+        mock_client = MagicMock()
-        mock_model.generate_content.return_value = _make_gemini_response([])
+        mock_client.models.generate_content.return_value = _make_gemini_response([])
 
         engine = GeminiEngine()
-        engine._model = mock_model
+        engine._client = mock_client
-        engine._generation_config = MagicMock()
+        engine._model_name = "gemini-2.5-flash"
 
         result = engine.extract_maintenance(_make_pdf_bytes())
 
@@ -223,12 +217,12 @@ class TestExtractMaintenance:
 
         schedule = [{"serviceName": "Brake Fluid Replacement"}]
 
-        mock_model = MagicMock()
+        mock_client = MagicMock()
-        mock_model.generate_content.return_value = _make_gemini_response(schedule)
+        mock_client.models.generate_content.return_value = _make_gemini_response(schedule)
 
         engine = GeminiEngine()
-        engine._model = mock_model
+        engine._client = mock_client
-        engine._generation_config = MagicMock()
+        engine._model_name = "gemini-2.5-flash"
 
         result = engine.extract_maintenance(_make_pdf_bytes())
 
@@ -264,7 +258,8 @@ class TestErrorHandling:
         with (
             patch("app.engines.gemini_engine.settings") as mock_settings,
             patch.dict("sys.modules", {
-                "google.cloud.aiplatform": None,
+                "google": None,
+                "google.genai": None,
             }),
         ):
             mock_settings.google_vision_key_path = "/fake/creds.json"
@@ -283,12 +278,12 @@ class TestErrorHandling:
         mock_settings.vertex_ai_location = "us-central1"
         mock_settings.gemini_model = "gemini-2.5-flash"
 
-        mock_model = MagicMock()
+        mock_client = MagicMock()
-        mock_model.generate_content.side_effect = RuntimeError("API quota exceeded")
+        mock_client.models.generate_content.side_effect = RuntimeError("API quota exceeded")
 
         engine = GeminiEngine()
-        engine._model = mock_model
+        engine._client = mock_client
-        engine._generation_config = MagicMock()
+        engine._model_name = "gemini-2.5-flash"
 
         with pytest.raises(GeminiProcessingError, match="maintenance extraction failed"):
             engine.extract_maintenance(_make_pdf_bytes())
@@ -307,12 +302,12 @@ class TestErrorHandling:
         mock_response = MagicMock()
         mock_response.text = "not valid json {{"
 
-        mock_model = MagicMock()
+        mock_client = MagicMock()
-        mock_model.generate_content.return_value = mock_response
+        mock_client.models.generate_content.return_value = mock_response
 
         engine = GeminiEngine()
-        engine._model = mock_model
+        engine._client = mock_client
-        engine._generation_config = MagicMock()
+        engine._model_name = "gemini-2.5-flash"
 
         with pytest.raises(GeminiProcessingError, match="invalid JSON"):
             engine.extract_maintenance(_make_pdf_bytes())
@@ -322,32 +317,32 @@ class TestErrorHandling:
 
 
 class TestLazyInitialization:
-    """Verify the model is not created until first use."""
+    """Verify the client is not created until first use."""
 
-    def test_model_is_none_after_construction(self):
+    def test_client_is_none_after_construction(self):
-        """GeminiEngine should not initialize the model in __init__."""
+        """GeminiEngine should not initialize the client in __init__."""
         engine = GeminiEngine()
-        assert engine._model is None
+        assert engine._client is None
 
     @patch("app.engines.gemini_engine.settings")
     @patch("app.engines.gemini_engine.os.path.isfile", return_value=True)
-    def test_model_reused_on_second_call(self, mock_isfile, mock_settings):
+    def test_client_reused_on_second_call(self, mock_isfile, mock_settings):
-        """Once initialized, the same model instance is reused."""
+        """Once initialized, the same client instance is reused."""
         mock_settings.google_vision_key_path = "/fake/creds.json"
         mock_settings.vertex_ai_project = "test-project"
         mock_settings.vertex_ai_location = "us-central1"
         mock_settings.gemini_model = "gemini-2.5-flash"
 
         schedule = [{"serviceName": "Oil Change", "intervalMiles": 5000}]
-        mock_model = MagicMock()
+        mock_client = MagicMock()
-        mock_model.generate_content.return_value = _make_gemini_response(schedule)
+        mock_client.models.generate_content.return_value = _make_gemini_response(schedule)
 
         engine = GeminiEngine()
-        engine._model = mock_model
+        engine._client = mock_client
-        engine._generation_config = MagicMock()
+        engine._model_name = "gemini-2.5-flash"
 
         engine.extract_maintenance(_make_pdf_bytes())
         engine.extract_maintenance(_make_pdf_bytes())
 
-        # Model's generate_content should have been called twice
+        # Client's generate_content should have been called twice
-        assert mock_model.generate_content.call_count == 2
+        assert mock_client.models.generate_content.call_count == 2